diff --git "a/wandb/run-20211106_211610-dtkf2u0m/files/output.log" "b/wandb/run-20211106_211610-dtkf2u0m/files/output.log"
--- "a/wandb/run-20211106_211610-dtkf2u0m/files/output.log"
+++ "b/wandb/run-20211106_211610-dtkf2u0m/files/output.log"
@@ -33702,3 +33702,3760 @@ To disable this warning, you can either:
 huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
 To disable this warning, you can either:
 	- Avoid using `tokenizers` before the fork if possible
+11/07/2021 16:16:05 - WARNING - huggingface_hub.repository - The progress bars may be unreliable.19424, 'steps': 134996, 'loss/train': 1.2643868923187256}}}}████████████████████████| 51.4M/51.4M [00:21<00:00, 2.55MB/s]
+11/07/2021 16:16:05 - WARNING - huggingface_hub.repository - The progress bars may be unreliable.19424, 'steps': 134996, 'loss/train': 1.2643868923187256}}}}████████████████████████| 51.4M/51.4M [00:21<00:00, 2.55MB/s]
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+	- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+	- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+	- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+Upload file log/debug_0.log:   0%|▏                                                                                                                                                          | 32.0k/19.7M [00:00<?, ?B/s]
+Upload file log/debug_0.log:  30%|████████████████████████████████████████████                                                                                                       | 5.91M/19.7M [00:01<00:02, 6.13MB/s]
+Upload file log/debug_0.log:  30%|████████████████████████████████████████████                                                                                                       | 5.91M/19.7M [00:01<00:02, 6.13MB/s]
+Upload file log/debug_0.log:  30%|████████████████████████████████████████████                                                                                                       | 5.91M/19.7M [00:01<00:02, 6.13MB/s]
+Upload file log/debug_0.log:  30%|████████████████████████████████████████████                                                                                                       | 5.91M/19.7M [00:01<00:02, 6.13MB/s]
+Upload file log/debug_0.log:  30%|████████████████████████████████████████████                                                                                                       | 5.91M/19.7M [00:01<00:02, 6.13MB/s]
+Upload file log/debug_0.log:  30%|████████████████████████████████████████████                                                                                                       | 5.91M/19.7M [00:01<00:02, 6.13MB/s]
+Upload file log/debug_0.log:  30%|████████████████████████████████████████████                                                                                                       | 5.91M/19.7M [00:01<00:02, 6.13MB/s]
+Upload file log/debug_0.log:  30%|████████████████████████████████████████████                                                                                                       | 5.91M/19.7M [00:01<00:02, 6.13MB/s]
+Upload file log/debug_0.log: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████���████████████████████████████████████████| 19.7M/19.7M [00:18<00:00, 6.13MB/s]
+Upload file log/debug_0.log: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+Upload file log/debug_0.log: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+	- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+	- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
+11/07/2021 16:16:30 - INFO - __main__ - Step 135005: {'lr': 1.2559343144586816e-05, 'samples': 25920960, 'steps': 135004, 'loss/train': 1.5371102094650269}██████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+11/07/2021 16:16:33 - INFO - __main__ - Step 135010: {'lr': 1.2551040192076784e-05, 'samples': 25921920, 'steps': 135009, 'loss/train': 1.3162504434585571}██████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+11/07/2021 16:16:33 - INFO - __main__ - Step 135010: {'lr': 1.2551040192076784e-05, 'samples': 25921920, 'steps': 135009, 'loss/train': 1.3162504434585571}██████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+11/07/2021 16:16:35 - INFO - __main__ - Step 135016: {'lr': 1.2541080179770569e-05, 'samples': 25923072, 'steps': 135015, 'loss/train': 1.3807737827301025}██████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+11/07/2021 16:16:38 - INFO - __main__ - Step 135022: {'lr': 1.2531124019296125e-05, 'samples': 25924224, 'steps': 135021, 'loss/train': 0.755631148815155}}██████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+11/07/2021 16:16:41 - INFO - __main__ - Step 135026: {'lr': 1.2524488718966054e-05, 'samples': 25924992, 'steps': 135025, 'loss/train': 2.6583170890808105}██████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+11/07/2021 16:16:43 - INFO - __main__ - Step 135032: {'lr': 1.251453897856994e-05, 'samples': 25926144, 'steps': 135031, 'loss/train': 1.965047836303711}5}██████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+11/07/2021 16:16:43 - INFO - __main__ - Step 135032: {'lr': 1.251453897856994e-05, 'samples': 25926144, 'steps': 135031, 'loss/train': 1.965047836303711}5}██████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+11/07/2021 16:16:46 - INFO - __main__ - Step 135037: {'lr': 1.2506250470933262e-05, 'samples': 25927104, 'steps': 135036, 'loss/train': 1.5263806581497192}██████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+11/07/2021 16:16:49 - INFO - __main__ - Step 135042: {'lr': 1.2497964638572085e-05, 'samples': 25928064, 'steps': 135041, 'loss/train': 1.3597644567489624}██████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+11/07/2021 16:16:49 - INFO - __main__ - Step 135042: {'lr': 1.2497964638572085e-05, 'samples': 25928064, 'steps': 135041, 'loss/train': 1.3597644567489624}██████████████████████████| 19.7M/19.7M [00:20<00:00, 1.03MB/s]
+/home/leandro/miniconda3/envs/codeparrot/lib/python3.8/site-packages/accelerate/accelerator.py:435: FutureWarning: Non-finite norm encountered in torch.nn.utils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:16:54 - INFO - __main__ - Step 135052: {'lr': 1.2481401000049463e-05, 'samples': 25929984, 'steps': 135051, 'loss/train': 0.9704720973968506}utils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:16:57 - INFO - __main__ - Step 135058: {'lr': 1.2471467953954486e-05, 'samples': 25931136, 'steps': 135057, 'loss/train': 1.213752269744873}}utils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:16:59 - INFO - __main__ - Step 135062: {'lr': 1.2464848063748341e-05, 'samples': 25931904, 'steps': 135061, 'loss/train': 1.5271512269973755}utils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:16:59 - INFO - __main__ - Step 135062: {'lr': 1.2464848063748341e-05, 'samples': 25931904, 'steps': 135061, 'loss/train': 1.5271512269973755}utils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:02 - INFO - __main__ - Step 135069: {'lr': 1.2453267376558774e-05, 'samples': 25933248, 'steps': 135068, 'loss/train': 1.4826006889343262}utils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:02 - INFO - __main__ - Step 135069: {'lr': 1.2453267376558774e-05, 'samples': 25933248, 'steps': 135068, 'loss/train': 1.4826006889343262}utils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:07 - INFO - __main__ - Step 135078: {'lr': 1.2438385628148751e-05, 'samples': 25934976, 'steps': 135077, 'loss/train': 1.6811840534210205}utils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:09 - INFO - __main__ - Step 135082: {'lr': 1.2431774300793564e-05, 'samples': 25935744, 'steps': 135081, 'loss/train': 1.1169989109039307}utils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:10 - INFO - __main__ - Step 135086: {'lr': 1.2425164686151596e-05, 'samples': 25936512, 'steps': 135085, 'loss/train': 1.3258172273635864}utils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:12 - INFO - __main__ - Step 135090: {'lr': 1.2418556784270508e-05, 'samples': 25937280, 'steps': 135089, 'loss/train': 1.4914780855178833}utils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:15 - INFO - __main__ - Step 135095: {'lr': 1.2410299315561673e-05, 'samples': 25938240, 'steps': 135094, 'loss/train': 0.03190302103757858}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:17 - INFO - __main__ - Step 135099: {'lr': 1.240369526756674e-05, 'samples': 25939008, 'steps': 135098, 'loss/train': 1.312318205833435}58}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:19 - INFO - __main__ - Step 135103: {'lr': 1.239709293248742e-05, 'samples': 25939776, 'steps': 135102, 'loss/train': 1.0584796667099}5}58}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:21 - INFO - __main__ - Step 135107: {'lr': 1.2390492310371343e-05, 'samples': 25940544, 'steps': 135106, 'loss/train': 1.2546484470367432}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:23 - INFO - __main__ - Step 135111: {'lr': 1.2383893401266166e-05, 'samples': 25941312, 'steps': 135110, 'loss/train': 1.7120970487594604}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:25 - INFO - __main__ - Step 135116: {'lr': 1.2375647173878596e-05, 'samples': 25942272, 'steps': 135115, 'loss/train': 1.3404165506362915}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:25 - INFO - __main__ - Step 135116: {'lr': 1.2375647173878596e-05, 'samples': 25942272, 'steps': 135115, 'loss/train': 1.3404165506362915}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:29 - INFO - __main__ - Step 135123: {'lr': 1.2364106952491267e-05, 'samples': 25943616, 'steps': 135122, 'loss/train': 1.3878759145736694}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:31 - INFO - __main__ - Step 135128: {'lr': 1.2355867149451694e-05, 'samples': 25944576, 'steps': 135127, 'loss/train': 1.1555670499801636}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:34 - INFO - __main__ - Step 135133: {'lr': 1.2347630023381628e-05, 'samples': 25945536, 'steps': 135132, 'loss/train': 1.4416487216949463}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:34 - INFO - __main__ - Step 135133: {'lr': 1.2347630023381628e-05, 'samples': 25945536, 'steps': 135132, 'loss/train': 1.4416487216949463}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:37 - INFO - __main__ - Step 135140: {'lr': 1.2336102544368926e-05, 'samples': 25946880, 'steps': 135139, 'loss/train': 1.642526388168335}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:39 - INFO - __main__ - Step 135144: {'lr': 1.2329517769417237e-05, 'samples': 25947648, 'steps': 135143, 'loss/train': 1.4660699367523193}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:41 - INFO - __main__ - Step 135149: {'lr': 1.2321289210272868e-05, 'samples': 25948608, 'steps': 135148, 'loss/train': 1.03617262840271}3}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:43 - INFO - __main__ - Step 135153: {'lr': 1.2314708290651427e-05, 'samples': 25949376, 'steps': 135152, 'loss/train': 0.9059475660324097}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:43 - INFO - __main__ - Step 135153: {'lr': 1.2314708290651427e-05, 'samples': 25949376, 'steps': 135152, 'loss/train': 0.9059475660324097}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:46 - INFO - __main__ - Step 135160: {'lr': 1.2303195804589357e-05, 'samples': 25950720, 'steps': 135159, 'loss/train': 1.1294143199920654}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:49 - INFO - __main__ - Step 135164: {'lr': 1.2296619597364278e-05, 'samples': 25951488, 'steps': 135163, 'loss/train': 1.5126222372055054}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:49 - INFO - __main__ - Step 135164: {'lr': 1.2296619597364278e-05, 'samples': 25951488, 'steps': 135163, 'loss/train': 1.5126222372055054}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:53 - INFO - __main__ - Step 135173: {'lr': 1.2281829396850409e-05, 'samples': 25953216, 'steps': 135172, 'loss/train': 1.1601303815841675}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:55 - INFO - __main__ - Step 135177: {'lr': 1.2275258759283342e-05, 'samples': 25953984, 'steps': 135176, 'loss/train': 1.2369372844696045}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:57 - INFO - __main__ - Step 135181: {'lr': 1.22686898355579e-05, 'samples': 25954752, 'steps': 135180, 'loss/train': 1.1058834791183472}5}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:17:59 - INFO - __main__ - Step 135186: {'lr': 1.2260481091062925e-05, 'samples': 25955712, 'steps': 135185, 'loss/train': 1.222041368484497}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:01 - INFO - __main__ - Step 135190: {'lr': 1.2253916023654193e-05, 'samples': 25956480, 'steps': 135189, 'loss/train': 1.4673658609390259}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:04 - INFO - __main__ - Step 135194: {'lr': 1.2247352670240936e-05, 'samples': 25957248, 'steps': 135193, 'loss/train': 0.9565936923027039}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:06 - INFO - __main__ - Step 135198: {'lr': 1.2240791030870446e-05, 'samples': 25958016, 'steps': 135197, 'loss/train': 1.1788297891616821}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:07 - INFO - __main__ - Step 135202: {'lr': 1.2234231105590048e-05, 'samples': 25958784, 'steps': 135201, 'loss/train': 1.4920928478240967}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:09 - INFO - __main__ - Step 135206: {'lr': 1.2227672894447067e-05, 'samples': 25959552, 'steps': 135205, 'loss/train': 1.4767874479293823}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:09 - INFO - __main__ - Step 135206: {'lr': 1.2227672894447067e-05, 'samples': 25959552, 'steps': 135205, 'loss/train': 1.4767874479293823}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:13 - INFO - __main__ - Step 135212: {'lr': 1.2217838791843544e-05, 'samples': 25960704, 'steps': 135211, 'loss/train': 1.1375707387924194}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:16 - INFO - __main__ - Step 135218: {'lr': 1.2208008546315042e-05, 'samples': 25961856, 'steps': 135217, 'loss/train': 1.0495827198028564}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:16 - INFO - __main__ - Step 135218: {'lr': 1.2208008546315042e-05, 'samples': 25961856, 'steps': 135217, 'loss/train': 1.0495827198028564}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:19 - INFO - __main__ - Step 135225: {'lr': 1.21965448016588e-05, 'samples': 25963200, 'steps': 135224, 'loss/train': 1.4095314741134644}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:21 - INFO - __main__ - Step 135229: {'lr': 1.2189996447722896e-05, 'samples': 25963968, 'steps': 135228, 'loss/train': 1.1181578636169434}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:24 - INFO - __main__ - Step 135234: {'lr': 1.2181813416262732e-05, 'samples': 25964928, 'steps': 135233, 'loss/train': 1.901018738746643}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:24 - INFO - __main__ - Step 135234: {'lr': 1.2181813416262732e-05, 'samples': 25964928, 'steps': 135233, 'loss/train': 1.901018738746643}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:27 - INFO - __main__ - Step 135241: {'lr': 1.217036167284169e-05, 'samples': 25966272, 'steps': 135240, 'loss/train': 1.6389700174331665}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:29 - INFO - __main__ - Step 135245: {'lr': 1.2163820177014062e-05, 'samples': 25967040, 'steps': 135244, 'loss/train': 1.1361795663833618}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:32 - INFO - __main__ - Step 135251: {'lr': 1.215401114824674e-05, 'samples': 25968192, 'steps': 135250, 'loss/train': 1.0418249368667603}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:32 - INFO - __main__ - Step 135251: {'lr': 1.215401114824674e-05, 'samples': 25968192, 'steps': 135250, 'loss/train': 1.0418249368667603}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:36 - INFO - __main__ - Step 135258: {'lr': 1.214257215758552e-05, 'samples': 25969536, 'steps': 135257, 'loss/train': 1.634677529335022}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:37 - INFO - __main__ - Step 135262: {'lr': 1.2136037949324718e-05, 'samples': 25970304, 'steps': 135261, 'loss/train': 0.912204921245575}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:39 - INFO - __main__ - Step 135266: {'lr': 1.2129505455909184e-05, 'samples': 25971072, 'steps': 135265, 'loss/train': 1.0998179912567139}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:42 - INFO - __main__ - Step 135271: {'lr': 1.2121342250712708e-05, 'samples': 25972032, 'steps': 135270, 'loss/train': 1.73945951461792}9}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:44 - INFO - __main__ - Step 135275: {'lr': 1.2114813615871273e-05, 'samples': 25972800, 'steps': 135274, 'loss/train': 1.100616216659546}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:44 - INFO - __main__ - Step 135275: {'lr': 1.2114813615871273e-05, 'samples': 25972800, 'steps': 135274, 'loss/train': 1.100616216659546}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:47 - INFO - __main__ - Step 135282: {'lr': 1.2103392631641701e-05, 'samples': 25974144, 'steps': 135281, 'loss/train': 0.8816916346549988}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:50 - INFO - __main__ - Step 135287: {'lr': 1.2095238001524982e-05, 'samples': 25975104, 'steps': 135286, 'loss/train': 1.4689027070999146}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:52 - INFO - __main__ - Step 135291: {'lr': 1.2088716226959035e-05, 'samples': 25975872, 'steps': 135290, 'loss/train': 1.2265511751174927}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:52 - INFO - __main__ - Step 135291: {'lr': 1.2088716226959035e-05, 'samples': 25975872, 'steps': 135290, 'loss/train': 1.2265511751174927}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:55 - INFO - __main__ - Step 135298: {'lr': 1.2077307248664294e-05, 'samples': 25977216, 'steps': 135297, 'loss/train': 1.4168705940246582}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:58 - INFO - __main__ - Step 135303: {'lr': 1.2069161194567913e-05, 'samples': 25978176, 'steps': 135302, 'loss/train': 1.6688258647918701}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:18:58 - INFO - __main__ - Step 135303: {'lr': 1.2069161194567913e-05, 'samples': 25978176, 'steps': 135302, 'loss/train': 1.6688258647918701}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:02 - INFO - __main__ - Step 135311: {'lr': 1.2056133082865989e-05, 'samples': 25979712, 'steps': 135310, 'loss/train': 1.3409314155578613}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:03 - INFO - __main__ - Step 135315: {'lr': 1.2049621600123629e-05, 'samples': 25980480, 'steps': 135314, 'loss/train': 1.0495047569274902}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:05 - INFO - __main__ - Step 135319: {'lr': 1.2043111832849508e-05, 'samples': 25981248, 'steps': 135318, 'loss/train': 1.1149972677230835}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:08 - INFO - __main__ - Step 135324: {'lr': 1.2034977036205652e-05, 'samples': 25982208, 'steps': 135323, 'loss/train': 1.0600417852401733}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:10 - INFO - __main__ - Step 135328: {'lr': 1.2028471128906814e-05, 'samples': 25982976, 'steps': 135327, 'loss/train': 1.3204816579818726}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:10 - INFO - __main__ - Step 135328: {'lr': 1.2028471128906814e-05, 'samples': 25982976, 'steps': 135327, 'loss/train': 1.3204816579818726}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:13 - INFO - __main__ - Step 135335: {'lr': 1.2017089919374501e-05, 'samples': 25984320, 'steps': 135334, 'loss/train': 1.515115737915039}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:16 - INFO - __main__ - Step 135340: {'lr': 1.2008963700922471e-05, 'samples': 25985280, 'steps': 135339, 'loss/train': 1.35220468044281}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:18 - INFO - __main__ - Step 135345: {'lr': 1.2000840163347625e-05, 'samples': 25986240, 'steps': 135344, 'loss/train': 1.0330960750579834}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:18 - INFO - __main__ - Step 135345: {'lr': 1.2000840163347625e-05, 'samples': 25986240, 'steps': 135344, 'loss/train': 1.0330960750579834}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:21 - INFO - __main__ - Step 135352: {'lr': 1.1989471714790784e-05, 'samples': 25987584, 'steps': 135351, 'loss/train': 1.1773396730422974}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:24 - INFO - __main__ - Step 135356: {'lr': 1.1982977817821677e-05, 'samples': 25988352, 'steps': 135355, 'loss/train': 1.5463366508483887}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:24 - INFO - __main__ - Step 135356: {'lr': 1.1982977817821677e-05, 'samples': 25988352, 'steps': 135355, 'loss/train': 1.5463366508483887}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:28 - INFO - __main__ - Step 135364: {'lr': 1.1969995171776493e-05, 'samples': 25989888, 'steps': 135363, 'loss/train': 1.2307099103927612}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:29 - INFO - __main__ - Step 135368: {'lr': 1.1963506422794063e-05, 'samples': 25990656, 'steps': 135367, 'loss/train': 1.4296314716339111}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:31 - INFO - __main__ - Step 135372: {'lr': 1.1957019389900736e-05, 'samples': 25991424, 'steps': 135371, 'loss/train': 1.3153777122497559}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:34 - INFO - __main__ - Step 135377: {'lr': 1.194891301210571e-05, 'samples': 25992384, 'steps': 135376, 'loss/train': 1.1948007345199585}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:34 - INFO - __main__ - Step 135377: {'lr': 1.194891301210571e-05, 'samples': 25992384, 'steps': 135376, 'loss/train': 1.1948007345199585}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:37 - INFO - __main__ - Step 135384: {'lr': 1.1937568588223202e-05, 'samples': 25993728, 'steps': 135383, 'loss/train': 1.3618885278701782}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:40 - INFO - __main__ - Step 135388: {'lr': 1.1931088420153974e-05, 'samples': 25994496, 'steps': 135387, 'loss/train': 1.5567642450332642}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:42 - INFO - __main__ - Step 135393: {'lr': 1.1922990623651902e-05, 'samples': 25995456, 'steps': 135392, 'loss/train': 1.1002498865127563}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:44 - INFO - __main__ - Step 135397: {'lr': 1.1916514317374755e-05, 'samples': 25996224, 'steps': 135396, 'loss/train': 1.4598093032836914}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:44 - INFO - __main__ - Step 135397: {'lr': 1.1916514317374755e-05, 'samples': 25996224, 'steps': 135396, 'loss/train': 1.4598093032836914}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:47 - INFO - __main__ - Step 135404: {'lr': 1.1905184911572498e-05, 'samples': 25997568, 'steps': 135403, 'loss/train': 1.2446706295013428}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:47 - INFO - __main__ - Step 135404: {'lr': 1.1905184911572498e-05, 'samples': 25997568, 'steps': 135403, 'loss/train': 1.2446706295013428}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:51 - INFO - __main__ - Step 135411: {'lr': 1.1893860762580866e-05, 'samples': 25998912, 'steps': 135410, 'loss/train': 1.3163864612579346}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:54 - INFO - __main__ - Step 135416: {'lr': 1.188577530332366e-05, 'samples': 25999872, 'steps': 135415, 'loss/train': 1.8137965202331543}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:56 - INFO - __main__ - Step 135420: {'lr': 1.187930886714425e-05, 'samples': 26000640, 'steps': 135419, 'loss/train': 1.3275854587554932}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:58 - INFO - __main__ - Step 135424: {'lr': 1.1872844147661015e-05, 'samples': 26001408, 'steps': 135423, 'loss/train': 0.7496057748794556}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:19:59 - INFO - __main__ - Step 135428: {'lr': 1.1866381144920475e-05, 'samples': 26002176, 'steps': 135427, 'loss/train': 1.2649413347244263}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:02 - INFO - __main__ - Step 135432: {'lr': 1.1859919858969308e-05, 'samples': 26002944, 'steps': 135431, 'loss/train': 1.1875102519989014}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:04 - INFO - __main__ - Step 135437: {'lr': 1.185184566583633e-05, 'samples': 26003904, 'steps': 135436, 'loss/train': 1.4490883350372314}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:06 - INFO - __main__ - Step 135441: {'lr': 1.1845388242831462e-05, 'samples': 26004672, 'steps': 135440, 'loss/train': 1.3556212186813354}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:06 - INFO - __main__ - Step 135441: {'lr': 1.1845388242831462e-05, 'samples': 26004672, 'steps': 135440, 'loss/train': 1.3556212186813354}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:09 - INFO - __main__ - Step 135448: {'lr': 1.1834091883989513e-05, 'samples': 26006016, 'steps': 135447, 'loss/train': 1.3498060703277588}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:12 - INFO - __main__ - Step 135453: {'lr': 1.1826026275647189e-05, 'samples': 26006976, 'steps': 135452, 'loss/train': 1.3446846008300781}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:12 - INFO - __main__ - Step 135453: {'lr': 1.1826026275647189e-05, 'samples': 26006976, 'steps': 135452, 'loss/train': 1.3446846008300781}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:12 - INFO - __main__ - Step 135453: {'lr': 1.1826026275647189e-05, 'samples': 26006976, 'steps': 135452, 'loss/train': 1.3446846008300781}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:18 - INFO - __main__ - Step 135463: {'lr': 1.1809903107867198e-05, 'samples': 26008896, 'steps': 135462, 'loss/train': 1.443830966949463}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:20 - INFO - __main__ - Step 135468: {'lr': 1.1801845548611106e-05, 'samples': 26009856, 'steps': 135467, 'loss/train': 1.1193368434906006}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:21 - INFO - __main__ - Step 135472: {'lr': 1.1795401433113306e-05, 'samples': 26010624, 'steps': 135471, 'loss/train': 1.2926826477050781}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:24 - INFO - __main__ - Step 135476: {'lr': 1.1788959034916614e-05, 'samples': 26011392, 'steps': 135475, 'loss/train': 1.3303331136703491}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:26 - INFO - __main__ - Step 135481: {'lr': 1.1780908452196065e-05, 'samples': 26012352, 'steps': 135480, 'loss/train': 0.5687029957771301}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:26 - INFO - __main__ - Step 135481: {'lr': 1.1780908452196065e-05, 'samples': 26012352, 'steps': 135480, 'loss/train': 0.5687029957771301}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:30 - INFO - __main__ - Step 135489: {'lr': 1.1768033101448972e-05, 'samples': 26013888, 'steps': 135488, 'loss/train': 1.66291081905365}1}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:32 - INFO - __main__ - Step 135493: {'lr': 1.1761598002299756e-05, 'samples': 26014656, 'steps': 135492, 'loss/train': 1.3214752674102783}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:34 - INFO - __main__ - Step 135497: {'lr': 1.1755164620695314e-05, 'samples': 26015424, 'steps': 135496, 'loss/train': 1.5633397102355957}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:36 - INFO - __main__ - Step 135502: {'lr': 1.174712530905775e-05, 'samples': 26016384, 'steps': 135501, 'loss/train': 1.260170340538025}7}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:39 - INFO - __main__ - Step 135507: {'lr': 1.1739088681246873e-05, 'samples': 26017344, 'steps': 135506, 'loss/train': 1.1644716262817383}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:39 - INFO - __main__ - Step 135507: {'lr': 1.1739088681246873e-05, 'samples': 26017344, 'steps': 135506, 'loss/train': 1.1644716262817383}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:43 - INFO - __main__ - Step 135514: {'lr': 1.1727841911312902e-05, 'samples': 26018688, 'steps': 135513, 'loss/train': 0.7414166331291199}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:44 - INFO - __main__ - Step 135518: {'lr': 1.1721417547579134e-05, 'samples': 26019456, 'steps': 135517, 'loss/train': 1.3645615577697754}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:46 - INFO - __main__ - Step 135522: {'lr': 1.1714994901679765e-05, 'samples': 26020224, 'steps': 135521, 'loss/train': 1.3159338235855103}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:49 - INFO - __main__ - Step 135527: {'lr': 1.1706969010080754e-05, 'samples': 26021184, 'steps': 135526, 'loss/train': 0.5964181423187256}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:49 - INFO - __main__ - Step 135527: {'lr': 1.1706969010080754e-05, 'samples': 26021184, 'steps': 135526, 'loss/train': 0.5964181423187256}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:52 - INFO - __main__ - Step 135534: {'lr': 1.1695737271451263e-05, 'samples': 26022528, 'steps': 135533, 'loss/train': 1.5054157972335815}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:52 - INFO - __main__ - Step 135534: {'lr': 1.1695737271451263e-05, 'samples': 26022528, 'steps': 135533, 'loss/train': 1.5054157972335815}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:54 - INFO - __main__ - Step 135538: {'lr': 1.1689321497352552e-05, 'samples': 26023296, 'steps': 135537, 'loss/train': 0.49768102169036865}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:20:57 - INFO - __main__ - Step 135544: {'lr': 1.167970105759239e-05, 'samples': 26024448, 'steps': 135543, 'loss/train': 1.1947540044784546}5}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:00 - INFO - __main__ - Step 135549: {'lr': 1.1671686977497126e-05, 'samples': 26025408, 'steps': 135548, 'loss/train': 1.3611621856689453}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:02 - INFO - __main__ - Step 135554: {'lr': 1.166367558207837e-05, 'samples': 26026368, 'steps': 135553, 'loss/train': 1.467131495475769}3}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:04 - INFO - __main__ - Step 135558: {'lr': 1.1657268398771126e-05, 'samples': 26027136, 'steps': 135557, 'loss/train': 1.5029581785202026}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:04 - INFO - __main__ - Step 135558: {'lr': 1.1657268398771126e-05, 'samples': 26027136, 'steps': 135557, 'loss/train': 1.5029581785202026}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:08 - INFO - __main__ - Step 135565: {'lr': 1.1646059962663102e-05, 'samples': 26028480, 'steps': 135564, 'loss/train': 1.2950758934020996}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:10 - INFO - __main__ - Step 135570: {'lr': 1.1638057158815468e-05, 'samples': 26029440, 'steps': 135569, 'loss/train': 1.1915842294692993}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:12 - INFO - __main__ - Step 135574: {'lr': 1.1631656848972855e-05, 'samples': 26030208, 'steps': 135573, 'loss/train': 0.9294772744178772}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:12 - INFO - __main__ - Step 135574: {'lr': 1.1631656848972855e-05, 'samples': 26030208, 'steps': 135573, 'loss/train': 0.9294772744178772}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:16 - INFO - __main__ - Step 135581: {'lr': 1.1620460441872288e-05, 'samples': 26031552, 'steps': 135580, 'loss/train': 1.069351077079773}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:18 - INFO - __main__ - Step 135586: {'lr': 1.1612466230518898e-05, 'samples': 26032512, 'steps': 135585, 'loss/train': 1.6917062997817993}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:20 - INFO - __main__ - Step 135590: {'lr': 1.1606072794879213e-05, 'samples': 26033280, 'steps': 135589, 'loss/train': 0.8353034853935242}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:22 - INFO - __main__ - Step 135594: {'lr': 1.1599681077905543e-05, 'samples': 26034048, 'steps': 135593, 'loss/train': 1.4904969930648804}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:22 - INFO - __main__ - Step 135594: {'lr': 1.1599681077905543e-05, 'samples': 26034048, 'steps': 135593, 'loss/train': 1.4904969930648804}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:26 - INFO - __main__ - Step 135601: {'lr': 1.1588499708880318e-05, 'samples': 26035392, 'steps': 135600, 'loss/train': 0.8548798561096191}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:28 - INFO - __main__ - Step 135606: {'lr': 1.1580516239441314e-05, 'samples': 26036352, 'steps': 135605, 'loss/train': 1.2419028282165527}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:30 - INFO - __main__ - Step 135611: {'lr': 1.157253545570583e-05, 'samples': 26037312, 'steps': 135610, 'loss/train': 1.4274933338165283}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:30 - INFO - __main__ - Step 135611: {'lr': 1.157253545570583e-05, 'samples': 26037312, 'steps': 135610, 'loss/train': 1.4274933338165283}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:34 - INFO - __main__ - Step 135618: {'lr': 1.1561366870629198e-05, 'samples': 26038656, 'steps': 135617, 'loss/train': 1.3113760948181152}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:36 - INFO - __main__ - Step 135622: {'lr': 1.1554987185607102e-05, 'samples': 26039424, 'steps': 135621, 'loss/train': 1.2977855205535889}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:38 - INFO - __main__ - Step 135627: {'lr': 1.1547014996726806e-05, 'samples': 26040384, 'steps': 135626, 'loss/train': 0.9938121438026428}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:41 - INFO - __main__ - Step 135632: {'lr': 1.1539045493927369e-05, 'samples': 26041344, 'steps': 135631, 'loss/train': 1.256483554840088}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:43 - INFO - __main__ - Step 135636: {'lr': 1.1532671825726455e-05, 'samples': 26042112, 'steps': 135635, 'loss/train': 1.5087766647338867}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:43 - INFO - __main__ - Step 135636: {'lr': 1.1532671825726455e-05, 'samples': 26042112, 'steps': 135635, 'loss/train': 1.5087766647338867}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:43 - INFO - __main__ - Step 135636: {'lr': 1.1532671825726455e-05, 'samples': 26042112, 'steps': 135635, 'loss/train': 1.5087766647338867}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:48 - INFO - __main__ - Step 135646: {'lr': 1.1516745176803427e-05, 'samples': 26044032, 'steps': 135645, 'loss/train': 0.8111945986747742}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:50 - INFO - __main__ - Step 135650: {'lr': 1.1510377525986492e-05, 'samples': 26044800, 'steps': 135649, 'loss/train': 1.3009686470031738}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:52 - INFO - __main__ - Step 135655: {'lr': 1.1502420380314988e-05, 'samples': 26045760, 'steps': 135654, 'loss/train': 1.1155062913894653}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:55 - INFO - __main__ - Step 135660: {'lr': 1.1494465921226688e-05, 'samples': 26046720, 'steps': 135659, 'loss/train': 1.9881243705749512}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:55 - INFO - __main__ - Step 135660: {'lr': 1.1494465921226688e-05, 'samples': 26046720, 'steps': 135659, 'loss/train': 1.9881243705749512}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:21:58 - INFO - __main__ - Step 135667: {'lr': 1.1483334192133515e-05, 'samples': 26048064, 'steps': 135666, 'loss/train': 0.9575623273849487}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:00 - INFO - __main__ - Step 135671: {'lr': 1.1476975568446928e-05, 'samples': 26048832, 'steps': 135670, 'loss/train': 1.1622428894042969}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:02 - INFO - __main__ - Step 135675: {'lr': 1.1470618664357057e-05, 'samples': 26049600, 'steps': 135674, 'loss/train': 1.3294038772583008}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:04 - INFO - __main__ - Step 135680: {'lr': 1.1462674952497421e-05, 'samples': 26050560, 'steps': 135679, 'loss/train': 1.201867699623108}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:07 - INFO - __main__ - Step 135685: {'lr': 1.1454733927668687e-05, 'samples': 26051520, 'steps': 135684, 'loss/train': 1.332000970840454}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:07 - INFO - __main__ - Step 135685: {'lr': 1.1454733927668687e-05, 'samples': 26051520, 'steps': 135684, 'loss/train': 1.332000970840454}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:10 - INFO - __main__ - Step 135692: {'lr': 1.1443621007290822e-05, 'samples': 26052864, 'steps': 135691, 'loss/train': 0.8679123520851135}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:12 - INFO - __main__ - Step 135696: {'lr': 1.1437273131835234e-05, 'samples': 26053632, 'steps': 135695, 'loss/train': 1.215234398841858}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:15 - INFO - __main__ - Step 135701: {'lr': 1.1429340706106516e-05, 'samples': 26054592, 'steps': 135700, 'loss/train': 1.2984423637390137}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:17 - INFO - __main__ - Step 135706: {'lr': 1.1421410967784234e-05, 'samples': 26055552, 'steps': 135705, 'loss/train': 1.42076575756073}7}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:17 - INFO - __main__ - Step 135706: {'lr': 1.1421410967784234e-05, 'samples': 26055552, 'steps': 135705, 'loss/train': 1.42076575756073}7}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:20 - INFO - __main__ - Step 135712: {'lr': 1.1411898829299772e-05, 'samples': 26056704, 'steps': 135711, 'loss/train': 1.030491828918457}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:23 - INFO - __main__ - Step 135717: {'lr': 1.140397500358606e-05, 'samples': 26057664, 'steps': 135716, 'loss/train': 1.4215322732925415}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:26 - INFO - __main__ - Step 135722: {'lr': 1.139605386556447e-05, 'samples': 26058624, 'steps': 135721, 'loss/train': 1.0979763269424438}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:26 - INFO - __main__ - Step 135722: {'lr': 1.139605386556447e-05, 'samples': 26058624, 'steps': 135721, 'loss/train': 1.0979763269424438}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:30 - INFO - __main__ - Step 135731: {'lr': 1.1381802590394785e-05, 'samples': 26060352, 'steps': 135730, 'loss/train': 1.2585958242416382}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:32 - INFO - __main__ - Step 135735: {'lr': 1.137547148574805e-05, 'samples': 26061120, 'steps': 135734, 'loss/train': 1.2821917533874512}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:34 - INFO - __main__ - Step 135739: {'lr': 1.136914210142978e-05, 'samples': 26061888, 'steps': 135738, 'loss/train': 1.5278267860412598}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:34 - INFO - __main__ - Step 135739: {'lr': 1.136914210142978e-05, 'samples': 26061888, 'steps': 135738, 'loss/train': 1.5278267860412598}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:38 - INFO - __main__ - Step 135746: {'lr': 1.1358069818550532e-05, 'samples': 26063232, 'steps': 135745, 'loss/train': 1.1413494348526}8}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:40 - INFO - __main__ - Step 135751: {'lr': 1.1350164270902214e-05, 'samples': 26064192, 'steps': 135750, 'loss/train': 1.212560772895813}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:40 - INFO - __main__ - Step 135751: {'lr': 1.1350164270902214e-05, 'samples': 26064192, 'steps': 135750, 'loss/train': 1.212560772895813}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:45 - INFO - __main__ - Step 135759: {'lr': 1.1337520986362509e-05, 'samples': 26065728, 'steps': 135758, 'loss/train': 0.9574289321899414}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:47 - INFO - __main__ - Step 135763: {'lr': 1.1331201924972972e-05, 'samples': 26066496, 'steps': 135762, 'loss/train': 0.9706791639328003}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:48 - INFO - __main__ - Step 135767: {'lr': 1.1324884584231087e-05, 'samples': 26067264, 'steps': 135766, 'loss/train': 1.542677640914917}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:50 - INFO - __main__ - Step 135771: {'lr': 1.1318568964182403e-05, 'samples': 26068032, 'steps': 135770, 'loss/train': 1.6192647218704224}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:53 - INFO - __main__ - Step 135776: {'lr': 1.131067685891568e-05, 'samples': 26068992, 'steps': 135775, 'loss/train': 1.256069302558899}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:53 - INFO - __main__ - Step 135776: {'lr': 1.131067685891568e-05, 'samples': 26068992, 'steps': 135775, 'loss/train': 1.256069302558899}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:57 - INFO - __main__ - Step 135784: {'lr': 1.129805508311163e-05, 'samples': 26070528, 'steps': 135783, 'loss/train': 1.0306817293167114}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:22:58 - INFO - __main__ - Step 135788: {'lr': 1.129174677651676e-05, 'samples': 26071296, 'steps': 135787, 'loss/train': 1.2430380582809448}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:00 - INFO - __main__ - Step 135792: {'lr': 1.128544019085398e-05, 'samples': 26072064, 'steps': 135791, 'loss/train': 1.8879588842391968}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:03 - INFO - __main__ - Step 135797: {'lr': 1.1277559378905567e-05, 'samples': 26073024, 'steps': 135796, 'loss/train': 0.9223677515983582}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:05 - INFO - __main__ - Step 135802: {'lr': 1.1269681256073277e-05, 'samples': 26073984, 'steps': 135801, 'loss/train': 1.40491783618927}2}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:05 - INFO - __main__ - Step 135802: {'lr': 1.1269681256073277e-05, 'samples': 26073984, 'steps': 135801, 'loss/train': 1.40491783618927}2}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:09 - INFO - __main__ - Step 135809: {'lr': 1.125865640199228e-05, 'samples': 26075328, 'steps': 135808, 'loss/train': 1.377108097076416}2}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:11 - INFO - __main__ - Step 135813: {'lr': 1.1252358851968624e-05, 'samples': 26076096, 'steps': 135812, 'loss/train': 1.2557326555252075}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:11 - INFO - __main__ - Step 135813: {'lr': 1.1252358851968624e-05, 'samples': 26076096, 'steps': 135812, 'loss/train': 1.2557326555252075}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:11 - INFO - __main__ - Step 135813: {'lr': 1.1252358851968624e-05, 'samples': 26076096, 'steps': 135812, 'loss/train': 1.2557326555252075}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:16 - INFO - __main__ - Step 135824: {'lr': 1.1235049464560276e-05, 'samples': 26078208, 'steps': 135823, 'loss/train': 1.323904275894165}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:18 - INFO - __main__ - Step 135829: {'lr': 1.1227185864488205e-05, 'samples': 26079168, 'steps': 135828, 'loss/train': 0.7227658033370972}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:18 - INFO - __main__ - Step 135829: {'lr': 1.1227185864488205e-05, 'samples': 26079168, 'steps': 135828, 'loss/train': 0.7227658033370972}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:18 - INFO - __main__ - Step 135829: {'lr': 1.1227185864488205e-05, 'samples': 26079168, 'steps': 135828, 'loss/train': 0.7227658033370972}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:24 - INFO - __main__ - Step 135839: {'lr': 1.1211466733483556e-05, 'samples': 26081088, 'steps': 135838, 'loss/train': 0.7188524007797241}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:24 - INFO - __main__ - Step 135839: {'lr': 1.1211466733483556e-05, 'samples': 26081088, 'steps': 135838, 'loss/train': 0.7188524007797241}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:24 - INFO - __main__ - Step 135839: {'lr': 1.1211466733483556e-05, 'samples': 26081088, 'steps': 135838, 'loss/train': 0.7188524007797241}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:30 - INFO - __main__ - Step 135851: {'lr': 1.1192617978804748e-05, 'samples': 26083392, 'steps': 135850, 'loss/train': 1.073644757270813}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:33 - INFO - __main__ - Step 135857: {'lr': 1.1183199411946648e-05, 'samples': 26084544, 'steps': 135856, 'loss/train': 1.3370330333709717}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:35 - INFO - __main__ - Step 135861: {'lr': 1.117692251950847e-05, 'samples': 26085312, 'steps': 135860, 'loss/train': 1.2704914808273315}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:35 - INFO - __main__ - Step 135861: {'lr': 1.117692251950847e-05, 'samples': 26085312, 'steps': 135860, 'loss/train': 1.2704914808273315}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:38 - INFO - __main__ - Step 135868: {'lr': 1.116594210075364e-05, 'samples': 26086656, 'steps': 135867, 'loss/train': 1.453705906867981}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:40 - INFO - __main__ - Step 135872: {'lr': 1.1159669943264006e-05, 'samples': 26087424, 'steps': 135871, 'loss/train': 1.8249189853668213}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:43 - INFO - __main__ - Step 135877: {'lr': 1.1151832167807018e-05, 'samples': 26088384, 'steps': 135876, 'loss/train': 1.4932277202606201}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:45 - INFO - __main__ - Step 135881: {'lr': 1.1145563884620625e-05, 'samples': 26089152, 'steps': 135880, 'loss/train': 1.3200455904006958}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:47 - INFO - __main__ - Step 135885: {'lr': 1.1139297323420144e-05, 'samples': 26089920, 'steps': 135884, 'loss/train': 1.5805184841156006}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:48 - INFO - __main__ - Step 135889: {'lr': 1.113303248425082e-05, 'samples': 26090688, 'steps': 135888, 'loss/train': 1.1650985479354858}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:51 - INFO - __main__ - Step 135893: {'lr': 1.112676936715784e-05, 'samples': 26091456, 'steps': 135892, 'loss/train': 1.4934967756271362}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:51 - INFO - __main__ - Step 135893: {'lr': 1.112676936715784e-05, 'samples': 26091456, 'steps': 135892, 'loss/train': 1.4934967756271362}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:55 - INFO - __main__ - Step 135901: {'lr': 1.1114248299381346e-05, 'samples': 26092992, 'steps': 135900, 'loss/train': 1.298383116722107}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:56 - INFO - __main__ - Step 135905: {'lr': 1.1107990348788178e-05, 'samples': 26093760, 'steps': 135904, 'loss/train': 1.2547963857650757}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:23:59 - INFO - __main__ - Step 135909: {'lr': 1.1101734120451818e-05, 'samples': 26094528, 'steps': 135908, 'loss/train': 1.6352767944335938}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:01 - INFO - __main__ - Step 135914: {'lr': 1.10939162570238e-05, 'samples': 26095488, 'steps': 135913, 'loss/train': 1.4364128112792969}8}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:01 - INFO - __main__ - Step 135914: {'lr': 1.10939162570238e-05, 'samples': 26095488, 'steps': 135913, 'loss/train': 1.4364128112792969}8}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:05 - INFO - __main__ - Step 135922: {'lr': 1.1081413273240109e-05, 'samples': 26097024, 'steps': 135921, 'loss/train': 1.2697639465332031}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:07 - INFO - __main__ - Step 135926: {'lr': 1.1075164364998418e-05, 'samples': 26097792, 'steps': 135925, 'loss/train': 1.026249885559082}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:09 - INFO - __main__ - Step 135930: {'lr': 1.106891717925021e-05, 'samples': 26098560, 'steps': 135929, 'loss/train': 1.629459261894226}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:11 - INFO - __main__ - Step 135935: {'lr': 1.106111061939008e-05, 'samples': 26099520, 'steps': 135934, 'loss/train': 1.083873987197876}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:13 - INFO - __main__ - Step 135939: {'lr': 1.1054867309416932e-05, 'samples': 26100288, 'steps': 135938, 'loss/train': 1.5653325319290161}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:15 - INFO - __main__ - Step 135943: {'lr': 1.1048625722083622e-05, 'samples': 26101056, 'steps': 135942, 'loss/train': 0.9213332533836365}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:15 - INFO - __main__ - Step 135943: {'lr': 1.1048625722083622e-05, 'samples': 26101056, 'steps': 135942, 'loss/train': 0.9213332533836365}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:19 - INFO - __main__ - Step 135950: {'lr': 1.1037707089487759e-05, 'samples': 26102400, 'steps': 135949, 'loss/train': 1.3396425247192383}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:21 - INFO - __main__ - Step 135955: {'lr': 1.1029911296372569e-05, 'samples': 26103360, 'steps': 135954, 'loss/train': 1.3660430908203125}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:24 - INFO - __main__ - Step 135960: {'lr': 1.1022118195163272e-05, 'samples': 26104320, 'steps': 135959, 'loss/train': 0.9425734877586365}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:24 - INFO - __main__ - Step 135960: {'lr': 1.1022118195163272e-05, 'samples': 26104320, 'steps': 135959, 'loss/train': 0.9425734877586365}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:27 - INFO - __main__ - Step 135967: {'lr': 1.1011212376039193e-05, 'samples': 26105664, 'steps': 135966, 'loss/train': 1.5484520196914673}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:29 - INFO - __main__ - Step 135971: {'lr': 1.1004982848443951e-05, 'samples': 26106432, 'steps': 135970, 'loss/train': 1.5143083333969116}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:31 - INFO - __main__ - Step 135976: {'lr': 1.0997198361923172e-05, 'samples': 26107392, 'steps': 135975, 'loss/train': 1.4317946434020996}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:31 - INFO - __main__ - Step 135976: {'lr': 1.0997198361923172e-05, 'samples': 26107392, 'steps': 135975, 'loss/train': 1.4317946434020996}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:35 - INFO - __main__ - Step 135984: {'lr': 1.098474878345701e-05, 'samples': 26108928, 'steps': 135983, 'loss/train': 1.6894962787628174}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:38 - INFO - __main__ - Step 135988: {'lr': 1.097852657891954e-05, 'samples': 26109696, 'steps': 135987, 'loss/train': 1.3710263967514038}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:39 - INFO - __main__ - Step 135992: {'lr': 1.0972306097572244e-05, 'samples': 26110464, 'steps': 135991, 'loss/train': 1.6659480333328247}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:41 - INFO - __main__ - Step 135996: {'lr': 1.0966087339460001e-05, 'samples': 26111232, 'steps': 135995, 'loss/train': 0.9373067021369934}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:41 - INFO - __main__ - Step 135996: {'lr': 1.0966087339460001e-05, 'samples': 26111232, 'steps': 135995, 'loss/train': 0.9373067021369934}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:46 - INFO - __main__ - Step 136004: {'lr': 1.0953654993119982e-05, 'samples': 26112768, 'steps': 136003, 'loss/train': 0.8742798566818237}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:47 - INFO - __main__ - Step 136008: {'lr': 1.09474414049818e-05, 'samples': 26113536, 'steps': 136007, 'loss/train': 1.2537329196929932}7}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:50 - INFO - __main__ - Step 136013: {'lr': 1.0939676843365748e-05, 'samples': 26114496, 'steps': 136012, 'loss/train': 1.3012884855270386}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:50 - INFO - __main__ - Step 136013: {'lr': 1.0939676843365748e-05, 'samples': 26114496, 'steps': 136012, 'loss/train': 1.3012884855270386}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:54 - INFO - __main__ - Step 136020: {'lr': 1.09288109812323e-05, 'samples': 26115840, 'steps': 136019, 'loss/train': 3.1832971572875977}6}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:55 - INFO - __main__ - Step 136024: {'lr': 1.0922604287019994e-05, 'samples': 26116608, 'steps': 136023, 'loss/train': 1.2824125289916992}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:24:57 - INFO - __main__ - Step 136028: {'lr': 1.0916399316401094e-05, 'samples': 26117376, 'steps': 136027, 'loss/train': 0.7951229810714722}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:00 - INFO - __main__ - Step 136033: {'lr': 1.0908645526998745e-05, 'samples': 26118336, 'steps': 136032, 'loss/train': 1.3458746671676636}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:02 - INFO - __main__ - Step 136037: {'lr': 1.0902444434628427e-05, 'samples': 26119104, 'steps': 136036, 'loss/train': 1.2778856754302979}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:02 - INFO - __main__ - Step 136037: {'lr': 1.0902444434628427e-05, 'samples': 26119104, 'steps': 136036, 'loss/train': 1.2778856754302979}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:05 - INFO - __main__ - Step 136044: {'lr': 1.089159667075354e-05, 'samples': 26120448, 'steps': 136043, 'loss/train': 1.2844810485839844}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:07 - INFO - __main__ - Step 136049: {'lr': 1.0883851500128495e-05, 'samples': 26121408, 'steps': 136048, 'loss/train': 1.4941964149475098}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:07 - INFO - __main__ - Step 136049: {'lr': 1.0883851500128495e-05, 'samples': 26121408, 'steps': 136048, 'loss/train': 1.4941964149475098}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:12 - INFO - __main__ - Step 136057: {'lr': 1.0871464829751126e-05, 'samples': 26122944, 'steps': 136056, 'loss/train': 1.0672423839569092}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:13 - INFO - __main__ - Step 136061: {'lr': 1.0865274080483185e-05, 'samples': 26123712, 'steps': 136060, 'loss/train': 1.5550107955932617}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:15 - INFO - __main__ - Step 136065: {'lr': 1.0859085055221901e-05, 'samples': 26124480, 'steps': 136064, 'loss/train': 0.9805464744567871}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:18 - INFO - __main__ - Step 136070: {'lr': 1.0851351198097715e-05, 'samples': 26125440, 'steps': 136069, 'loss/train': 1.3764426708221436}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:20 - INFO - __main__ - Step 136074: {'lr': 1.0845166052014604e-05, 'samples': 26126208, 'steps': 136073, 'loss/train': 1.1261621713638306}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:20 - INFO - __main__ - Step 136074: {'lr': 1.0845166052014604e-05, 'samples': 26126208, 'steps': 136073, 'loss/train': 1.1261621713638306}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:23 - INFO - __main__ - Step 136081: {'lr': 1.0834346195135874e-05, 'samples': 26127552, 'steps': 136080, 'loss/train': 1.2307252883911133}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:23 - INFO - __main__ - Step 136081: {'lr': 1.0834346195135874e-05, 'samples': 26127552, 'steps': 136080, 'loss/train': 1.2307252883911133}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:28 - INFO - __main__ - Step 136090: {'lr': 1.0820442709644273e-05, 'samples': 26129280, 'steps': 136089, 'loss/train': 1.2049833536148071}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:30 - INFO - __main__ - Step 136094: {'lr': 1.0814266184765053e-05, 'samples': 26130048, 'steps': 136093, 'loss/train': 1.4530023336410522}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:31 - INFO - __main__ - Step 136098: {'lr': 1.0808091384260226e-05, 'samples': 26130816, 'steps': 136097, 'loss/train': 1.3678991794586182}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:31 - INFO - __main__ - Step 136098: {'lr': 1.0808091384260226e-05, 'samples': 26130816, 'steps': 136097, 'loss/train': 1.3678991794586182}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:35 - INFO - __main__ - Step 136105: {'lr': 1.0797289632786589e-05, 'samples': 26132160, 'steps': 136104, 'loss/train': 1.3305031061172485}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:37 - INFO - __main__ - Step 136109: {'lr': 1.0791119574540903e-05, 'samples': 26132928, 'steps': 136108, 'loss/train': 1.277653455734253}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:39 - INFO - __main__ - Step 136114: {'lr': 1.0783409426875168e-05, 'samples': 26133888, 'steps': 136113, 'loss/train': 1.2405847311019897}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:42 - INFO - __main__ - Step 136118: {'lr': 1.0777243248909913e-05, 'samples': 26134656, 'steps': 136117, 'loss/train': 1.736167311668396}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:44 - INFO - __main__ - Step 136122: {'lr': 1.0771078795585976e-05, 'samples': 26135424, 'steps': 136121, 'loss/train': 1.1501095294952393}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:45 - INFO - __main__ - Step 136126: {'lr': 1.0764916066947795e-05, 'samples': 26136192, 'steps': 136125, 'loss/train': 1.384940266609192}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:47 - INFO - __main__ - Step 136130: {'lr': 1.0758755063039776e-05, 'samples': 26136960, 'steps': 136129, 'loss/train': 1.4684898853302002}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:50 - INFO - __main__ - Step 136135: {'lr': 1.0751056233624324e-05, 'samples': 26137920, 'steps': 136134, 'loss/train': 1.3990440368652344}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:50 - INFO - __main__ - Step 136135: {'lr': 1.0751056233624324e-05, 'samples': 26137920, 'steps': 136134, 'loss/train': 1.3990440368652344}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:53 - INFO - __main__ - Step 136142: {'lr': 1.0740282400141106e-05, 'samples': 26139264, 'steps': 136141, 'loss/train': 1.33977472782135}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:55 - INFO - __main__ - Step 136146: {'lr': 1.0734128295597916e-05, 'samples': 26140032, 'steps': 136145, 'loss/train': 1.4058430194854736}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:25:58 - INFO - __main__ - Step 136151: {'lr': 1.0726438090638141e-05, 'samples': 26140992, 'steps': 136150, 'loss/train': 2.1677463054656982}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:00 - INFO - __main__ - Step 136156: {'lr': 1.0718750581002717e-05, 'samples': 26141952, 'steps': 136155, 'loss/train': 1.5306342840194702}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:00 - INFO - __main__ - Step 136156: {'lr': 1.0718750581002717e-05, 'samples': 26141952, 'steps': 136155, 'loss/train': 1.5306342840194702}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:03 - INFO - __main__ - Step 136162: {'lr': 1.0709529127390315e-05, 'samples': 26143104, 'steps': 136161, 'loss/train': 1.1773700714111328}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:06 - INFO - __main__ - Step 136167: {'lr': 1.0701847547773258e-05, 'samples': 26144064, 'steps': 136166, 'loss/train': 1.016560673713684}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:08 - INFO - __main__ - Step 136172: {'lr': 1.0694168663757609e-05, 'samples': 26145024, 'steps': 136171, 'loss/train': 1.5928759574890137}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:08 - INFO - __main__ - Step 136172: {'lr': 1.0694168663757609e-05, 'samples': 26145024, 'steps': 136171, 'loss/train': 1.5928759574890137}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:11 - INFO - __main__ - Step 136178: {'lr': 1.068495756125437e-05, 'samples': 26146176, 'steps': 136177, 'loss/train': 1.0415385961532593}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:14 - INFO - __main__ - Step 136184: {'lr': 1.0675750340691126e-05, 'samples': 26147328, 'steps': 136183, 'loss/train': 1.2124711275100708}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:16 - INFO - __main__ - Step 136188: {'lr': 1.0669614350356016e-05, 'samples': 26148096, 'steps': 136187, 'loss/train': 0.8751630783081055}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:16 - INFO - __main__ - Step 136188: {'lr': 1.0669614350356016e-05, 'samples': 26148096, 'steps': 136187, 'loss/train': 0.8751630783081055}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:19 - INFO - __main__ - Step 136195: {'lr': 1.0658880519081454e-05, 'samples': 26149440, 'steps': 136194, 'loss/train': 1.4309544563293457}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:22 - INFO - __main__ - Step 136200: {'lr': 1.0651216732031094e-05, 'samples': 26150400, 'steps': 136199, 'loss/train': 1.272822618484497}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:24 - INFO - __main__ - Step 136204: {'lr': 1.0645087643630286e-05, 'samples': 26151168, 'steps': 136203, 'loss/train': 1.1969388723373413}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:26 - INFO - __main__ - Step 136209: {'lr': 1.063742870975118e-05, 'samples': 26152128, 'steps': 136208, 'loss/train': 1.6625349521636963}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:26 - INFO - __main__ - Step 136209: {'lr': 1.063742870975118e-05, 'samples': 26152128, 'steps': 136208, 'loss/train': 1.6625349521636963}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:30 - INFO - __main__ - Step 136215: {'lr': 1.062824154825548e-05, 'samples': 26153280, 'steps': 136214, 'loss/train': 1.3268036842346191}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:32 - INFO - __main__ - Step 136220: {'lr': 1.0620588546409626e-05, 'samples': 26154240, 'steps': 136219, 'loss/train': 0.5520867109298706}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:34 - INFO - __main__ - Step 136224: {'lr': 1.0614468086420858e-05, 'samples': 26155008, 'steps': 136223, 'loss/train': 1.2932932376861572}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:36 - INFO - __main__ - Step 136228: {'lr': 1.0608349352246915e-05, 'samples': 26155776, 'steps': 136227, 'loss/train': 1.2698994874954224}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:38 - INFO - __main__ - Step 136232: {'lr': 1.06022323439319e-05, 'samples': 26156544, 'steps': 136231, 'loss/train': 1.1000622510910034}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:40 - INFO - __main__ - Step 136236: {'lr': 1.0596117061519916e-05, 'samples': 26157312, 'steps': 136235, 'loss/train': 1.8093523979187012}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:42 - INFO - __main__ - Step 136241: {'lr': 1.0588475385623298e-05, 'samples': 26158272, 'steps': 136240, 'loss/train': 1.4455759525299072}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:42 - INFO - __main__ - Step 136241: {'lr': 1.0588475385623298e-05, 'samples': 26158272, 'steps': 136240, 'loss/train': 1.4455759525299072}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:46 - INFO - __main__ - Step 136248: {'lr': 1.057778157014308e-05, 'samples': 26159616, 'steps': 136247, 'loss/train': 1.1525229215621948}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:48 - INFO - __main__ - Step 136252: {'lr': 1.0571673191783982e-05, 'samples': 26160384, 'steps': 136251, 'loss/train': 1.26215660572052}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:50 - INFO - __main__ - Step 136257: {'lr': 1.0564040146201299e-05, 'samples': 26161344, 'steps': 136256, 'loss/train': 1.2612123489379883}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:50 - INFO - __main__ - Step 136257: {'lr': 1.0564040146201299e-05, 'samples': 26161344, 'steps': 136256, 'loss/train': 1.2612123489379883}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:50 - INFO - __main__ - Step 136257: {'lr': 1.0564040146201299e-05, 'samples': 26161344, 'steps': 136256, 'loss/train': 1.2612123489379883}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:55 - INFO - __main__ - Step 136267: {'lr': 1.0548782146582947e-05, 'samples': 26163264, 'steps': 136266, 'loss/train': 1.252924919128418}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:26:58 - INFO - __main__ - Step 136272: {'lr': 1.0541157192719253e-05, 'samples': 26164224, 'steps': 136271, 'loss/train': 1.296242594718933}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:00 - INFO - __main__ - Step 136277: {'lr': 1.0533534936266992e-05, 'samples': 26165184, 'steps': 136276, 'loss/train': 0.4198194146156311}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:00 - INFO - __main__ - Step 136277: {'lr': 1.0533534936266992e-05, 'samples': 26165184, 'steps': 136276, 'loss/train': 0.4198194146156311}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:04 - INFO - __main__ - Step 136284: {'lr': 1.0522868309048612e-05, 'samples': 26166528, 'steps': 136283, 'loss/train': 2.0329666137695312}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:06 - INFO - __main__ - Step 136288: {'lr': 1.0516775467383555e-05, 'samples': 26167296, 'steps': 136287, 'loss/train': 1.3324081897735596}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:06 - INFO - __main__ - Step 136288: {'lr': 1.0516775467383555e-05, 'samples': 26167296, 'steps': 136287, 'loss/train': 1.3324081897735596}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:06 - INFO - __main__ - Step 136288: {'lr': 1.0516775467383555e-05, 'samples': 26167296, 'steps': 136287, 'loss/train': 1.3324081897735596}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:12 - INFO - __main__ - Step 136299: {'lr': 1.050002905530098e-05, 'samples': 26169408, 'steps': 136298, 'loss/train': 1.235713243484497}6}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:14 - INFO - __main__ - Step 136303: {'lr': 1.0493942688309027e-05, 'samples': 26170176, 'steps': 136302, 'loss/train': 1.435092806816101}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:14 - INFO - __main__ - Step 136303: {'lr': 1.0493942688309027e-05, 'samples': 26170176, 'steps': 136302, 'loss/train': 1.435092806816101}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:18 - INFO - __main__ - Step 136311: {'lr': 1.0481775134420224e-05, 'samples': 26171712, 'steps': 136310, 'loss/train': 1.6078777313232422}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:18 - INFO - __main__ - Step 136311: {'lr': 1.0481775134420224e-05, 'samples': 26171712, 'steps': 136310, 'loss/train': 1.6078777313232422}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:18 - INFO - __main__ - Step 136311: {'lr': 1.0481775134420224e-05, 'samples': 26171712, 'steps': 136310, 'loss/train': 1.6078777313232422}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:24 - INFO - __main__ - Step 136322: {'lr': 1.046505602587075e-05, 'samples': 26173824, 'steps': 136321, 'loss/train': 1.030562400817871}2}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:26 - INFO - __main__ - Step 136328: {'lr': 1.0455942016526498e-05, 'samples': 26174976, 'steps': 136327, 'loss/train': 1.6355170011520386}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:26 - INFO - __main__ - Step 136328: {'lr': 1.0455942016526498e-05, 'samples': 26174976, 'steps': 136327, 'loss/train': 1.6355170011520386}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:29 - INFO - __main__ - Step 136335: {'lr': 1.0445313916671223e-05, 'samples': 26176320, 'steps': 136334, 'loss/train': 1.2063486576080322}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:31 - INFO - __main__ - Step 136339: {'lr': 1.043924309141106e-05, 'samples': 26177088, 'steps': 136338, 'loss/train': 0.9453842043876648}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:34 - INFO - __main__ - Step 136344: {'lr': 1.0431656988544536e-05, 'samples': 26178048, 'steps': 136343, 'loss/train': 0.9136726260185242}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:36 - INFO - __main__ - Step 136349: {'lr': 1.0424073584322985e-05, 'samples': 26179008, 'steps': 136348, 'loss/train': 1.2770488262176514}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:36 - INFO - __main__ - Step 136349: {'lr': 1.0424073584322985e-05, 'samples': 26179008, 'steps': 136348, 'loss/train': 1.2770488262176514}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:40 - INFO - __main__ - Step 136356: {'lr': 1.0413461352298953e-05, 'samples': 26180352, 'steps': 136355, 'loss/train': 1.2805461883544922}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:42 - INFO - __main__ - Step 136360: {'lr': 1.0407399594686701e-05, 'samples': 26181120, 'steps': 136359, 'loss/train': 0.9106904864311218}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:45 - INFO - __main__ - Step 136364: {'lr': 1.0401339564382162e-05, 'samples': 26181888, 'steps': 136363, 'loss/train': 1.6893272399902344}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:46 - INFO - __main__ - Step 136368: {'lr': 1.0395281261428913e-05, 'samples': 26182656, 'steps': 136367, 'loss/train': 0.8270352482795715}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:48 - INFO - __main__ - Step 136372: {'lr': 1.0389224685870697e-05, 'samples': 26183424, 'steps': 136371, 'loss/train': 1.564154863357544}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:48 - INFO - __main__ - Step 136372: {'lr': 1.0389224685870697e-05, 'samples': 26183424, 'steps': 136371, 'loss/train': 1.564154863357544}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:52 - INFO - __main__ - Step 136380: {'lr': 1.0377116717113998e-05, 'samples': 26184960, 'steps': 136379, 'loss/train': 1.059626579284668}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:54 - INFO - __main__ - Step 136384: {'lr': 1.0371065324002781e-05, 'samples': 26185728, 'steps': 136383, 'loss/train': 1.062570333480835}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:56 - INFO - __main__ - Step 136389: {'lr': 1.036350351201365e-05, 'samples': 26186688, 'steps': 136388, 'loss/train': 0.9593856930732727}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:59 - INFO - __main__ - Step 136393: {'lr': 1.0357456005995358e-05, 'samples': 26187456, 'steps': 136392, 'loss/train': 1.8399873971939087}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:27:59 - INFO - __main__ - Step 136393: {'lr': 1.0357456005995358e-05, 'samples': 26187456, 'steps': 136392, 'loss/train': 1.8399873971939087}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:02 - INFO - __main__ - Step 136399: {'lr': 1.0348387986358537e-05, 'samples': 26188608, 'steps': 136398, 'loss/train': 1.2196071147918701}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:05 - INFO - __main__ - Step 136403: {'lr': 1.0342344799591314e-05, 'samples': 26189376, 'steps': 136402, 'loss/train': 1.2978260517120361}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:06 - INFO - __main__ - Step 136407: {'lr': 1.0336303340600706e-05, 'samples': 26190144, 'steps': 136406, 'loss/train': 1.4462324380874634}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:08 - INFO - __main__ - Step 136412: {'lr': 1.0328753946614728e-05, 'samples': 26191104, 'steps': 136411, 'loss/train': 1.0401928424835205}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:11 - INFO - __main__ - Step 136417: {'lr': 1.0321207252432908e-05, 'samples': 26192064, 'steps': 136416, 'loss/train': 1.3749572038650513}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:11 - INFO - __main__ - Step 136417: {'lr': 1.0321207252432908e-05, 'samples': 26192064, 'steps': 136416, 'loss/train': 1.3749572038650513}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:14 - INFO - __main__ - Step 136424: {'lr': 1.0310646416411078e-05, 'samples': 26193408, 'steps': 136423, 'loss/train': 1.6040234565734863}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:16 - INFO - __main__ - Step 136428: {'lr': 1.030461402896224e-05, 'samples': 26194176, 'steps': 136427, 'loss/train': 1.2040830850601196}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:19 - INFO - __main__ - Step 136433: {'lr': 1.0297075974724734e-05, 'samples': 26195136, 'steps': 136432, 'loss/train': 0.7440516352653503}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:21 - INFO - __main__ - Step 136437: {'lr': 1.0291047475446514e-05, 'samples': 26195904, 'steps': 136436, 'loss/train': 1.5176669359207153}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:21 - INFO - __main__ - Step 136437: {'lr': 1.0291047475446514e-05, 'samples': 26195904, 'steps': 136436, 'loss/train': 1.5176669359207153}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:25 - INFO - __main__ - Step 136445: {'lr': 1.0278995661373024e-05, 'samples': 26197440, 'steps': 136444, 'loss/train': 1.1327580213546753}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:26 - INFO - __main__ - Step 136449: {'lr': 1.0272972346664605e-05, 'samples': 26198208, 'steps': 136448, 'loss/train': 1.5088077783584595}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:29 - INFO - __main__ - Step 136453: {'lr': 1.0266950760233012e-05, 'samples': 26198976, 'steps': 136452, 'loss/train': 0.1866033524274826}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:29 - INFO - __main__ - Step 136453: {'lr': 1.0266950760233012e-05, 'samples': 26198976, 'steps': 136452, 'loss/train': 0.1866033524274826}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:32 - INFO - __main__ - Step 136460: {'lr': 1.0256417142774227e-05, 'samples': 26200320, 'steps': 136459, 'loss/train': 1.5169953107833862}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:34 - INFO - __main__ - Step 136464: {'lr': 1.0250400309327612e-05, 'samples': 26201088, 'steps': 136463, 'loss/train': 1.384292721748352}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:36 - INFO - __main__ - Step 136469: {'lr': 1.0242881698142442e-05, 'samples': 26202048, 'steps': 136468, 'loss/train': 1.0140832662582397}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:39 - INFO - __main__ - Step 136474: {'lr': 1.023536578772874e-05, 'samples': 26203008, 'steps': 136473, 'loss/train': 1.2961151599884033}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:39 - INFO - __main__ - Step 136474: {'lr': 1.023536578772874e-05, 'samples': 26203008, 'steps': 136473, 'loss/train': 1.2961151599884033}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:42 - INFO - __main__ - Step 136481: {'lr': 1.022484805060675e-05, 'samples': 26204352, 'steps': 136480, 'loss/train': 1.5816162824630737}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:44 - INFO - __main__ - Step 136485: {'lr': 1.021884029195122e-05, 'samples': 26205120, 'steps': 136484, 'loss/train': 0.14996911585330963}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:47 - INFO - __main__ - Step 136491: {'lr': 1.020983189523217e-05, 'samples': 26206272, 'steps': 136490, 'loss/train': 5.664328575134277}3}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:47 - INFO - __main__ - Step 136491: {'lr': 1.020983189523217e-05, 'samples': 26206272, 'steps': 136490, 'loss/train': 5.664328575134277}3}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:50 - INFO - __main__ - Step 136498: {'lr': 1.0199327015151805e-05, 'samples': 26207616, 'steps': 136497, 'loss/train': 1.226137638092041}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:52 - INFO - __main__ - Step 136502: {'lr': 1.0193326603630204e-05, 'samples': 26208384, 'steps': 136501, 'loss/train': 1.6160025596618652}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:55 - INFO - __main__ - Step 136507: {'lr': 1.018582852042993e-05, 'samples': 26209344, 'steps': 136506, 'loss/train': 1.4041714668273926}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:55 - INFO - __main__ - Step 136507: {'lr': 1.018582852042993e-05, 'samples': 26209344, 'steps': 136506, 'loss/train': 1.4041714668273926}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:28:59 - INFO - __main__ - Step 136514: {'lr': 1.0175335742344249e-05, 'samples': 26210688, 'steps': 136513, 'loss/train': 1.0539764165878296}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:00 - INFO - __main__ - Step 136518: {'lr': 1.0169342246485996e-05, 'samples': 26211456, 'steps': 136517, 'loss/train': 0.5389702320098877}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:03 - INFO - __main__ - Step 136522: {'lr': 1.0163350479651668e-05, 'samples': 26212224, 'steps': 136521, 'loss/train': 1.7544224262237549}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:05 - INFO - __main__ - Step 136527: {'lr': 1.0155863202614412e-05, 'samples': 26213184, 'steps': 136526, 'loss/train': 1.2075698375701904}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:07 - INFO - __main__ - Step 136531: {'lr': 1.014987532624173e-05, 'samples': 26213952, 'steps': 136530, 'loss/train': 1.1045253276824951}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:09 - INFO - __main__ - Step 136535: {'lr': 1.0143889179033305e-05, 'samples': 26214720, 'steps': 136534, 'loss/train': 1.407990574836731}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:11 - INFO - __main__ - Step 136539: {'lr': 1.0137904761032324e-05, 'samples': 26215488, 'steps': 136538, 'loss/train': 1.2892552614212036}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:13 - INFO - __main__ - Step 136543: {'lr': 1.0131922072281835e-05, 'samples': 26216256, 'steps': 136542, 'loss/train': 1.3135340213775635}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:15 - INFO - __main__ - Step 136548: {'lr': 1.0124446143167987e-05, 'samples': 26217216, 'steps': 136547, 'loss/train': 1.4059066772460938}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:17 - INFO - __main__ - Step 136552: {'lr': 1.0118467345388932e-05, 'samples': 26217984, 'steps': 136551, 'loss/train': 1.8892319202423096}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:17 - INFO - __main__ - Step 136552: {'lr': 1.0118467345388932e-05, 'samples': 26217984, 'steps': 136551, 'loss/train': 1.8892319202423096}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:20 - INFO - __main__ - Step 136559: {'lr': 1.0108008610647728e-05, 'samples': 26219328, 'steps': 136558, 'loss/train': 1.3859730958938599}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:20 - INFO - __main__ - Step 136559: {'lr': 1.0108008610647728e-05, 'samples': 26219328, 'steps': 136558, 'loss/train': 1.3859730958938599}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:25 - INFO - __main__ - Step 136567: {'lr': 1.0096062256454764e-05, 'samples': 26220864, 'steps': 136566, 'loss/train': 0.9614728689193726}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:26 - INFO - __main__ - Step 136571: {'lr': 1.0090091673665036e-05, 'samples': 26221632, 'steps': 136570, 'loss/train': 1.4724904298782349}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:28 - INFO - __main__ - Step 136575: {'lr': 1.0084122820470554e-05, 'samples': 26222400, 'steps': 136574, 'loss/train': 1.2285979986190796}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:31 - INFO - __main__ - Step 136580: {'lr': 1.0076664186286333e-05, 'samples': 26223360, 'steps': 136579, 'loss/train': 1.7208986282348633}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:31 - INFO - __main__ - Step 136580: {'lr': 1.0076664186286333e-05, 'samples': 26223360, 'steps': 136579, 'loss/train': 1.7208986282348633}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:34 - INFO - __main__ - Step 136587: {'lr': 1.0066226638888815e-05, 'samples': 26224704, 'steps': 136586, 'loss/train': 1.3722970485687256}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:37 - INFO - __main__ - Step 136592: {'lr': 1.0058774491190859e-05, 'samples': 26225664, 'steps': 136591, 'loss/train': 1.7409542798995972}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:39 - INFO - __main__ - Step 136596: {'lr': 1.0052814719077068e-05, 'samples': 26226432, 'steps': 136595, 'loss/train': 1.699560523033142}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:41 - INFO - __main__ - Step 136600: {'lr': 1.0046856676827282e-05, 'samples': 26227200, 'steps': 136599, 'loss/train': 1.0723334550857544}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:41 - INFO - __main__ - Step 136600: {'lr': 1.0046856676827282e-05, 'samples': 26227200, 'steps': 136599, 'loss/train': 1.0723334550857544}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:44 - INFO - __main__ - Step 136607: {'lr': 1.0036434265504574e-05, 'samples': 26228544, 'steps': 136606, 'loss/train': 1.3402577638626099}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:47 - INFO - __main__ - Step 136612: {'lr': 1.0028992929691338e-05, 'samples': 26229504, 'steps': 136611, 'loss/train': 1.1520626544952393}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:49 - INFO - __main__ - Step 136616: {'lr': 1.0023041807326883e-05, 'samples': 26230272, 'steps': 136615, 'loss/train': 0.8715529441833496}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:51 - INFO - __main__ - Step 136620: {'lr': 1.001709241504109e-05, 'samples': 26231040, 'steps': 136619, 'loss/train': 1.2316575050354004}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:52 - INFO - __main__ - Step 136624: {'lr': 1.0011144752876816e-05, 'samples': 26231808, 'steps': 136623, 'loss/train': 1.3872902393341064}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:54 - INFO - __main__ - Step 136628: {'lr': 1.0005198820876915e-05, 'samples': 26232576, 'steps': 136627, 'loss/train': 1.3159968852996826}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:57 - INFO - __main__ - Step 136633: {'lr': 9.997768838986065e-06, 'samples': 26233536, 'steps': 136632, 'loss/train': 1.3698956966400146}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:59 - INFO - __main__ - Step 136638: {'lr': 9.990341560627725e-06, 'samples': 26234496, 'steps': 136637, 'loss/train': 0.38158538937568665}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:29:59 - INFO - __main__ - Step 136638: {'lr': 9.990341560627725e-06, 'samples': 26234496, 'steps': 136637, 'loss/train': 0.38158538937568665}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:03 - INFO - __main__ - Step 136645: {'lr': 9.979947913020037e-06, 'samples': 26235840, 'steps': 136644, 'loss/train': 1.879041314125061}5}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:05 - INFO - __main__ - Step 136649: {'lr': 9.97401106508622e-06, 'samples': 26236608, 'steps': 136648, 'loss/train': 1.74509859085083}1}5}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:07 - INFO - __main__ - Step 136653: {'lr': 9.968075947584503e-06, 'samples': 26237376, 'steps': 136652, 'loss/train': 1.4575023651123047}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:09 - INFO - __main__ - Step 136658: {'lr': 9.960659484192724e-06, 'samples': 26238336, 'steps': 136657, 'loss/train': 1.720950961112976}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:11 - INFO - __main__ - Step 136662: {'lr': 9.954728260319679e-06, 'samples': 26239104, 'steps': 136661, 'loss/train': 0.7407291531562805}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:11 - INFO - __main__ - Step 136662: {'lr': 9.954728260319679e-06, 'samples': 26239104, 'steps': 136661, 'loss/train': 0.7407291531562805}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:14 - INFO - __main__ - Step 136669: {'lr': 9.944352782754324e-06, 'samples': 26240448, 'steps': 136668, 'loss/train': 1.562712550163269}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:17 - INFO - __main__ - Step 136674: {'lr': 9.9369449722983e-06, 'samples': 26241408, 'steps': 136673, 'loss/train': 1.2373487949371338}}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:19 - INFO - __main__ - Step 136679: {'lr': 9.929539866059933e-06, 'samples': 26242368, 'steps': 136678, 'loss/train': 0.9784823656082153}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:19 - INFO - __main__ - Step 136679: {'lr': 9.929539866059933e-06, 'samples': 26242368, 'steps': 136678, 'loss/train': 0.9784823656082153}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:23 - INFO - __main__ - Step 136686: {'lr': 9.91917726057079e-06, 'samples': 26243712, 'steps': 136685, 'loss/train': 1.4090861082077026}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:25 - INFO - __main__ - Step 136690: {'lr': 9.913258151592886e-06, 'samples': 26244480, 'steps': 136689, 'loss/train': 1.7253601551055908}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:27 - INFO - __main__ - Step 136695: {'lr': 9.905861699411572e-06, 'samples': 26245440, 'steps': 136694, 'loss/train': 1.670967698097229}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:29 - INFO - __main__ - Step 136699: {'lr': 9.899946484951405e-06, 'samples': 26246208, 'steps': 136698, 'loss/train': 1.1285231113433838}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:31 - INFO - __main__ - Step 136703: {'lr': 9.894033001457275e-06, 'samples': 26246976, 'steps': 136702, 'loss/train': 1.6566308736801147}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:33 - INFO - __main__ - Step 136707: {'lr': 9.888121248971815e-06, 'samples': 26247744, 'steps': 136706, 'loss/train': 1.4060145616531372}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:35 - INFO - __main__ - Step 136711: {'lr': 9.88221122753774e-06, 'samples': 26248512, 'steps': 136710, 'loss/train': 1.3455414772033691}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:37 - INFO - __main__ - Step 136715: {'lr': 9.876302937197546e-06, 'samples': 26249280, 'steps': 136714, 'loss/train': 1.1961108446121216}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:37 - INFO - __main__ - Step 136715: {'lr': 9.876302937197546e-06, 'samples': 26249280, 'steps': 136714, 'loss/train': 1.1961108446121216}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:41 - INFO - __main__ - Step 136723: {'lr': 9.864491549969323e-06, 'samples': 26250816, 'steps': 136722, 'loss/train': 0.4997015595436096}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:43 - INFO - __main__ - Step 136727: {'lr': 9.858588453166423e-06, 'samples': 26251584, 'steps': 136726, 'loss/train': 1.1653170585632324}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:45 - INFO - __main__ - Step 136731: {'lr': 9.852687087627765e-06, 'samples': 26252352, 'steps': 136730, 'loss/train': 1.700740098953247}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:45 - INFO - __main__ - Step 136731: {'lr': 9.852687087627765e-06, 'samples': 26252352, 'steps': 136730, 'loss/train': 1.700740098953247}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:48 - INFO - __main__ - Step 136737: {'lr': 9.843838285533258e-06, 'samples': 26253504, 'steps': 136736, 'loss/train': 1.6691244840621948}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:51 - INFO - __main__ - Step 136743: {'lr': 9.83499337902255e-06, 'samples': 26254656, 'steps': 136742, 'loss/train': 1.0089002847671509}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:53 - INFO - __main__ - Step 136747: {'lr': 9.829098938966136e-06, 'samples': 26255424, 'steps': 136746, 'loss/train': 1.1803332567214966}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:55 - INFO - __main__ - Step 136751: {'lr': 9.823206230386517e-06, 'samples': 26256192, 'steps': 136750, 'loss/train': 1.3117533922195435}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:55 - INFO - __main__ - Step 136751: {'lr': 9.823206230386517e-06, 'samples': 26256192, 'steps': 136750, 'loss/train': 1.3117533922195435}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:30:58 - INFO - __main__ - Step 136758: {'lr': 9.812898156866174e-06, 'samples': 26257536, 'steps': 136757, 'loss/train': 1.271177887916565}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:01 - INFO - __main__ - Step 136764: {'lr': 9.804066886028084e-06, 'samples': 26258688, 'steps': 136763, 'loss/train': 1.2065004110336304}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:01 - INFO - __main__ - Step 136764: {'lr': 9.804066886028084e-06, 'samples': 26258688, 'steps': 136763, 'loss/train': 1.2065004110336304}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:05 - INFO - __main__ - Step 136771: {'lr': 9.793768661127522e-06, 'samples': 26260032, 'steps': 136770, 'loss/train': 1.3637733459472656}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:07 - INFO - __main__ - Step 136775: {'lr': 9.7878863423006e-06, 'samples': 26260800, 'steps': 136774, 'loss/train': 1.412257194519043}56}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:09 - INFO - __main__ - Step 136779: {'lr': 9.782005755247653e-06, 'samples': 26261568, 'steps': 136778, 'loss/train': 1.0732187032699585}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:09 - INFO - __main__ - Step 136779: {'lr': 9.782005755247653e-06, 'samples': 26261568, 'steps': 136778, 'loss/train': 1.0732187032699585}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:13 - INFO - __main__ - Step 136788: {'lr': 9.768780766396535e-06, 'samples': 26263296, 'steps': 136787, 'loss/train': 1.5497233867645264}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:13 - INFO - __main__ - Step 136788: {'lr': 9.768780766396535e-06, 'samples': 26263296, 'steps': 136787, 'loss/train': 1.5497233867645264}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:17 - INFO - __main__ - Step 136795: {'lr': 9.758500725622982e-06, 'samples': 26264640, 'steps': 136794, 'loss/train': 1.0499242544174194}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:20 - INFO - __main__ - Step 136801: {'lr': 9.749693483809551e-06, 'samples': 26265792, 'steps': 136800, 'loss/train': 1.0906379222869873}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:20 - INFO - __main__ - Step 136801: {'lr': 9.749693483809551e-06, 'samples': 26265792, 'steps': 136800, 'loss/train': 1.0906379222869873}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:23 - INFO - __main__ - Step 136808: {'lr': 9.739423293885774e-06, 'samples': 26267136, 'steps': 136807, 'loss/train': 1.4170857667922974}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:25 - INFO - __main__ - Step 136812: {'lr': 9.73355699558437e-06, 'samples': 26267904, 'steps': 136811, 'loss/train': 1.1027125120162964}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:27 - INFO - __main__ - Step 136817: {'lr': 9.726226558570444e-06, 'samples': 26268864, 'steps': 136816, 'loss/train': 0.9975262880325317}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:30 - INFO - __main__ - Step 136821: {'lr': 9.720364157701168e-06, 'samples': 26269632, 'steps': 136820, 'loss/train': 1.5189919471740723}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:30 - INFO - __main__ - Step 136821: {'lr': 9.720364157701168e-06, 'samples': 26269632, 'steps': 136820, 'loss/train': 1.5189919471740723}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:33 - INFO - __main__ - Step 136828: {'lr': 9.710109124457717e-06, 'samples': 26270976, 'steps': 136827, 'loss/train': 1.2332651615142822}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:35 - INFO - __main__ - Step 136833: {'lr': 9.702787348826708e-06, 'samples': 26271936, 'steps': 136832, 'loss/train': 1.3734630346298218}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:35 - INFO - __main__ - Step 136833: {'lr': 9.702787348826708e-06, 'samples': 26271936, 'steps': 136832, 'loss/train': 1.3734630346298218}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:39 - INFO - __main__ - Step 136841: {'lr': 9.691078138110571e-06, 'samples': 26273472, 'steps': 136840, 'loss/train': 1.4488166570663452}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:42 - INFO - __main__ - Step 136845: {'lr': 9.685226131439211e-06, 'samples': 26274240, 'steps': 136844, 'loss/train': 1.4099892377853394}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:43 - INFO - __main__ - Step 136849: {'lr': 9.679375857281959e-06, 'samples': 26275008, 'steps': 136848, 'loss/train': 1.7351144552230835}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:45 - INFO - __main__ - Step 136854: {'lr': 9.672065450997496e-06, 'samples': 26275968, 'steps': 136853, 'loss/train': 0.8688152432441711}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:47 - INFO - __main__ - Step 136858: {'lr': 9.666219075151027e-06, 'samples': 26276736, 'steps': 136857, 'loss/train': 1.0842281579971313}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:50 - INFO - __main__ - Step 136862: {'lr': 9.660374431955665e-06, 'samples': 26277504, 'steps': 136861, 'loss/train': 0.8730409741401672}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:51 - INFO - __main__ - Step 136866: {'lr': 9.654531521453513e-06, 'samples': 26278272, 'steps': 136865, 'loss/train': 0.8710697889328003}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:53 - INFO - __main__ - Step 136870: {'lr': 9.648690343686706e-06, 'samples': 26279040, 'steps': 136869, 'loss/train': 1.22221839427948}3}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:53 - INFO - __main__ - Step 136870: {'lr': 9.648690343686706e-06, 'samples': 26279040, 'steps': 136869, 'loss/train': 1.22221839427948}3}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:58 - INFO - __main__ - Step 136878: {'lr': 9.637013186527599e-06, 'samples': 26280576, 'steps': 136877, 'loss/train': 1.3844342231750488}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:31:59 - INFO - __main__ - Step 136882: {'lr': 9.631177207219427e-06, 'samples': 26281344, 'steps': 136881, 'loss/train': 1.1011008024215698}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:01 - INFO - __main__ - Step 136886: {'lr': 9.625342960815048e-06, 'samples': 26282112, 'steps': 136885, 'loss/train': 1.487064003944397}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:04 - INFO - __main__ - Step 136891: {'lr': 9.618052589769443e-06, 'samples': 26283072, 'steps': 136890, 'loss/train': 0.6772016286849976}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:04 - INFO - __main__ - Step 136891: {'lr': 9.618052589769443e-06, 'samples': 26283072, 'steps': 136890, 'loss/train': 0.6772016286849976}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:07 - INFO - __main__ - Step 136896: {'lr': 9.610764926533733e-06, 'samples': 26284032, 'steps': 136895, 'loss/train': 1.7044651508331299}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:09 - INFO - __main__ - Step 136902: {'lr': 9.602023305075564e-06, 'samples': 26285184, 'steps': 136901, 'loss/train': 1.291307806968689}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:09 - INFO - __main__ - Step 136902: {'lr': 9.602023305075564e-06, 'samples': 26285184, 'steps': 136901, 'loss/train': 1.291307806968689}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:13 - INFO - __main__ - Step 136909: {'lr': 9.591829675260288e-06, 'samples': 26286528, 'steps': 136908, 'loss/train': 1.3879331350326538}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:15 - INFO - __main__ - Step 136913: {'lr': 9.586007127055513e-06, 'samples': 26287296, 'steps': 136912, 'loss/train': 1.050858736038208}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:17 - INFO - __main__ - Step 136918: {'lr': 9.57873137915824e-06, 'samples': 26288256, 'steps': 136917, 'loss/train': 1.2654732465744019}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:20 - INFO - __main__ - Step 136923: {'lr': 9.571458339513783e-06, 'samples': 26289216, 'steps': 136922, 'loss/train': 1.3805075883865356}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:20 - INFO - __main__ - Step 136923: {'lr': 9.571458339513783e-06, 'samples': 26289216, 'steps': 136922, 'loss/train': 1.3805075883865356}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:23 - INFO - __main__ - Step 136930: {'lr': 9.56128063403236e-06, 'samples': 26290560, 'steps': 136929, 'loss/train': 1.1868278980255127}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:25 - INFO - __main__ - Step 136934: {'lr': 9.555467185749733e-06, 'samples': 26291328, 'steps': 136933, 'loss/train': 1.151458740234375}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:28 - INFO - __main__ - Step 136939: {'lr': 9.548202813064788e-06, 'samples': 26292288, 'steps': 136938, 'loss/train': 1.4144363403320312}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:30 - INFO - __main__ - Step 136943: {'lr': 9.54239326510259e-06, 'samples': 26293056, 'steps': 136942, 'loss/train': 1.0140916109085083}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:31 - INFO - __main__ - Step 136947: {'lr': 9.536585450684142e-06, 'samples': 26293824, 'steps': 136946, 'loss/train': 0.814092755317688}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:33 - INFO - __main__ - Step 136951: {'lr': 9.53077936985136e-06, 'samples': 26294592, 'steps': 136950, 'loss/train': 1.0558383464813232}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:36 - INFO - __main__ - Step 136956: {'lr': 9.523524206729001e-06, 'samples': 26295552, 'steps': 136955, 'loss/train': 1.188301920890808}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:38 - INFO - __main__ - Step 136960: {'lr': 9.517722026616993e-06, 'samples': 26296320, 'steps': 136959, 'loss/train': 1.228353500366211}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:38 - INFO - __main__ - Step 136960: {'lr': 9.517722026616993e-06, 'samples': 26296320, 'steps': 136959, 'loss/train': 1.228353500366211}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:41 - INFO - __main__ - Step 136967: {'lr': 9.507572383213898e-06, 'samples': 26297664, 'steps': 136966, 'loss/train': 0.07310379296541214}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:44 - INFO - __main__ - Step 136972: {'lr': 9.500325888778166e-06, 'samples': 26298624, 'steps': 136971, 'loss/train': 0.8118339776992798}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:46 - INFO - __main__ - Step 136976: {'lr': 9.49453064380365e-06, 'samples': 26299392, 'steps': 136975, 'loss/train': 1.3231821060180664}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:48 - INFO - __main__ - Step 136980: {'lr': 9.488737132718e-06, 'samples': 26300160, 'steps': 136979, 'loss/train': 0.6331982016563416}4}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:49 - INFO - __main__ - Step 136984: {'lr': 9.482945355562934e-06, 'samples': 26300928, 'steps': 136983, 'loss/train': 1.2228444814682007}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:52 - INFO - __main__ - Step 136988: {'lr': 9.477155312380249e-06, 'samples': 26301696, 'steps': 136987, 'loss/train': 1.1891615390777588}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:54 - INFO - __main__ - Step 136993: {'lr': 9.469920196864157e-06, 'samples': 26302656, 'steps': 136992, 'loss/train': 0.9952117800712585}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:56 - INFO - __main__ - Step 136997: {'lr': 9.464134055271911e-06, 'samples': 26303424, 'steps': 136996, 'loss/train': 1.687744379043579}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:58 - INFO - __main__ - Step 137001: {'lr': 9.458349647787662e-06, 'samples': 26304192, 'steps': 137000, 'loss/train': 1.232543706893921}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:32:58 - INFO - __main__ - Step 137001: {'lr': 9.458349647787662e-06, 'samples': 26304192, 'steps': 137000, 'loss/train': 1.232543706893921}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:01 - INFO - __main__ - Step 137008: {'lr': 9.448231107513e-06, 'samples': 26305536, 'steps': 137007, 'loss/train': 1.1073987483978271}1}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:04 - INFO - __main__ - Step 137013: {'lr': 9.441006830399834e-06, 'samples': 26306496, 'steps': 137012, 'loss/train': 1.5286551713943481}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:04 - INFO - __main__ - Step 137013: {'lr': 9.441006830399834e-06, 'samples': 26306496, 'steps': 137012, 'loss/train': 1.5286551713943481}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:04 - INFO - __main__ - Step 137013: {'lr': 9.441006830399834e-06, 'samples': 26306496, 'steps': 137012, 'loss/train': 1.5286551713943481}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:09 - INFO - __main__ - Step 137024: {'lr': 9.425122959376337e-06, 'samples': 26308608, 'steps': 137023, 'loss/train': 1.635650873184204}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:12 - INFO - __main__ - Step 137029: {'lr': 9.41790735392306e-06, 'samples': 26309568, 'steps': 137028, 'loss/train': 1.553197979927063}}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:14 - INFO - __main__ - Step 137033: {'lr': 9.412136820802942e-06, 'samples': 26310336, 'steps': 137032, 'loss/train': 1.002617597579956}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:16 - INFO - __main__ - Step 137037: {'lr': 9.40636802216574e-06, 'samples': 26311104, 'steps': 137036, 'loss/train': 1.4701272249221802}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:16 - INFO - __main__ - Step 137037: {'lr': 9.40636802216574e-06, 'samples': 26311104, 'steps': 137036, 'loss/train': 1.4701272249221802}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:19 - INFO - __main__ - Step 137044: {'lr': 9.396276798274983e-06, 'samples': 26312448, 'steps': 137043, 'loss/train': 0.9961363077163696}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:22 - INFO - __main__ - Step 137050: {'lr': 9.387631405869485e-06, 'samples': 26313600, 'steps': 137049, 'loss/train': 1.0666885375976562}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:24 - INFO - __main__ - Step 137054: {'lr': 9.381869979248198e-06, 'samples': 26314368, 'steps': 137053, 'loss/train': 1.0500967502593994}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:24 - INFO - __main__ - Step 137054: {'lr': 9.381869979248198e-06, 'samples': 26314368, 'steps': 137053, 'loss/train': 1.0500967502593994}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:27 - INFO - __main__ - Step 137061: {'lr': 9.371791656810402e-06, 'samples': 26315712, 'steps': 137060, 'loss/train': 0.027460573241114616}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:29 - INFO - __main__ - Step 137065: {'lr': 9.36603500071695e-06, 'samples': 26316480, 'steps': 137064, 'loss/train': 1.14303719997406}4616}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:32 - INFO - __main__ - Step 137070: {'lr': 9.358841620188879e-06, 'samples': 26317440, 'steps': 137069, 'loss/train': 1.5481892824172974}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:34 - INFO - __main__ - Step 137074: {'lr': 9.353088867488064e-06, 'samples': 26318208, 'steps': 137073, 'loss/train': 1.2786626815795898}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:34 - INFO - __main__ - Step 137074: {'lr': 9.353088867488064e-06, 'samples': 26318208, 'steps': 137073, 'loss/train': 1.2786626815795898}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:37 - INFO - __main__ - Step 137081: {'lr': 9.343025724910192e-06, 'samples': 26319552, 'steps': 137080, 'loss/train': 1.1728557348251343}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:40 - INFO - __main__ - Step 137086: {'lr': 9.335841019002644e-06, 'samples': 26320512, 'steps': 137085, 'loss/train': 1.3315519094467163}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:42 - INFO - __main__ - Step 137090: {'lr': 9.330095206184747e-06, 'samples': 26321280, 'steps': 137089, 'loss/train': 0.9570795297622681}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:44 - INFO - __main__ - Step 137094: {'lr': 9.324351128441155e-06, 'samples': 26322048, 'steps': 137093, 'loss/train': 1.6282659769058228}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:46 - INFO - __main__ - Step 137098: {'lr': 9.318608785813281e-06, 'samples': 26322816, 'steps': 137097, 'loss/train': 1.2872872352600098}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:48 - INFO - __main__ - Step 137103: {'lr': 9.311433297598066e-06, 'samples': 26323776, 'steps': 137102, 'loss/train': 1.327391266822815}}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:50 - INFO - __main__ - Step 137107: {'lr': 9.305694859131935e-06, 'samples': 26324544, 'steps': 137106, 'loss/train': 1.228474497795105}}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:50 - INFO - __main__ - Step 137107: {'lr': 9.305694859131935e-06, 'samples': 26324544, 'steps': 137106, 'loss/train': 1.228474497795105}}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:53 - INFO - __main__ - Step 137114: {'lr': 9.295656767286997e-06, 'samples': 26325888, 'steps': 137113, 'loss/train': 1.5352727174758911}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:56 - INFO - __main__ - Step 137118: {'lr': 9.289923100858576e-06, 'samples': 26326656, 'steps': 137117, 'loss/train': 0.9574427604675293}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:58 - INFO - __main__ - Step 137123: {'lr': 9.282758458183383e-06, 'samples': 26327616, 'steps': 137122, 'loss/train': 1.4515742063522339}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:33:58 - INFO - __main__ - Step 137123: {'lr': 9.282758458183383e-06, 'samples': 26327616, 'steps': 137122, 'loss/train': 1.4515742063522339}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:02 - INFO - __main__ - Step 137131: {'lr': 9.271300670037341e-06, 'samples': 26329152, 'steps': 137130, 'loss/train': 1.1173487901687622}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:04 - INFO - __main__ - Step 137135: {'lr': 9.265574379190955e-06, 'samples': 26329920, 'steps': 137134, 'loss/train': 1.1337287425994873}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:06 - INFO - __main__ - Step 137139: {'lr': 9.259849823884031e-06, 'samples': 26330688, 'steps': 137138, 'loss/train': 1.2814937829971313}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:08 - INFO - __main__ - Step 137144: {'lr': 9.252696570415691e-06, 'samples': 26331648, 'steps': 137143, 'loss/train': 1.2993767261505127}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:10 - INFO - __main__ - Step 137149: {'lr': 9.245546028872814e-06, 'samples': 26332608, 'steps': 137148, 'loss/train': 1.1279855966567993}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:10 - INFO - __main__ - Step 137149: {'lr': 9.245546028872814e-06, 'samples': 26332608, 'steps': 137148, 'loss/train': 1.1279855966567993}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:14 - INFO - __main__ - Step 137156: {'lr': 9.23553982690098e-06, 'samples': 26333952, 'steps': 137155, 'loss/train': 1.2479487657546997}}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:16 - INFO - __main__ - Step 137160: {'lr': 9.229824383853141e-06, 'samples': 26334720, 'steps': 137159, 'loss/train': 1.3949143886566162}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:18 - INFO - __main__ - Step 137164: {'lr': 9.224110676602504e-06, 'samples': 26335488, 'steps': 137163, 'loss/train': 1.046335220336914}}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:18 - INFO - __main__ - Step 137164: {'lr': 9.224110676602504e-06, 'samples': 26335488, 'steps': 137163, 'loss/train': 1.046335220336914}}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:22 - INFO - __main__ - Step 137172: {'lr': 9.212688469657643e-06, 'samples': 26337024, 'steps': 137171, 'loss/train': 1.1641908884048462}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:24 - INFO - __main__ - Step 137176: {'lr': 9.206979970045742e-06, 'samples': 26337792, 'steps': 137175, 'loss/train': 0.906486451625824}}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:25 - INFO - __main__ - Step 137180: {'lr': 9.201273206395743e-06, 'samples': 26338560, 'steps': 137179, 'loss/train': 1.2309901714324951}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:28 - INFO - __main__ - Step 137184: {'lr': 9.195568178748809e-06, 'samples': 26339328, 'steps': 137183, 'loss/train': 1.4686540365219116}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:30 - INFO - __main__ - Step 137189: {'lr': 9.188439335507087e-06, 'samples': 26340288, 'steps': 137188, 'loss/train': 1.2832690477371216}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:32 - INFO - __main__ - Step 137193: {'lr': 9.182738214017355e-06, 'samples': 26341056, 'steps': 137192, 'loss/train': 1.4074627161026}16}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:32 - INFO - __main__ - Step 137193: {'lr': 9.182738214017355e-06, 'samples': 26341056, 'steps': 137192, 'loss/train': 1.4074627161026}16}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:35 - INFO - __main__ - Step 137200: {'lr': 9.172765429013935e-06, 'samples': 26342400, 'steps': 137199, 'loss/train': 1.0672509670257568}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:38 - INFO - __main__ - Step 137205: {'lr': 9.165645266532573e-06, 'samples': 26343360, 'steps': 137204, 'loss/train': 1.0672193765640259}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:40 - INFO - __main__ - Step 137210: {'lr': 9.158527816957334e-06, 'samples': 26344320, 'steps': 137209, 'loss/train': 0.9108512997627258}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:40 - INFO - __main__ - Step 137210: {'lr': 9.158527816957334e-06, 'samples': 26344320, 'steps': 137209, 'loss/train': 0.9108512997627258}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:44 - INFO - __main__ - Step 137217: {'lr': 9.14856794538685e-06, 'samples': 26345664, 'steps': 137216, 'loss/train': 1.040697693824768}8}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:45 - INFO - __main__ - Step 137221: {'lr': 9.142878977716235e-06, 'samples': 26346432, 'steps': 137220, 'loss/train': 1.1989129781723022}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:45 - INFO - __main__ - Step 137221: {'lr': 9.142878977716235e-06, 'samples': 26346432, 'steps': 137220, 'loss/train': 1.1989129781723022}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:49 - INFO - __main__ - Step 137228: {'lr': 9.132927462587731e-06, 'samples': 26347776, 'steps': 137227, 'loss/train': 1.345660924911499}}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:51 - INFO - __main__ - Step 137232: {'lr': 9.127243270181884e-06, 'samples': 26348544, 'steps': 137231, 'loss/train': 1.4661513566970825}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:54 - INFO - __main__ - Step 137237: {'lr': 9.120140471684219e-06, 'samples': 26349504, 'steps': 137236, 'loss/train': 1.4018833637237549}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:56 - INFO - __main__ - Step 137242: {'lr': 9.113040386605209e-06, 'samples': 26350464, 'steps': 137241, 'loss/train': 0.9783901572227478}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:34:56 - INFO - __main__ - Step 137242: {'lr': 9.113040386605209e-06, 'samples': 26350464, 'steps': 137241, 'loss/train': 0.9783901572227478}6}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:00 - INFO - __main__ - Step 137247: {'lr': 9.105943015024904e-06, 'samples': 26351424, 'steps': 137246, 'loss/train': 0.47210144996643066}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:00 - INFO - __main__ - Step 137247: {'lr': 9.105943015024904e-06, 'samples': 26351424, 'steps': 137246, 'loss/train': 0.47210144996643066}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:04 - INFO - __main__ - Step 137256: {'lr': 9.093174584452107e-06, 'samples': 26353152, 'steps': 137255, 'loss/train': 1.2345809936523438}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:07 - INFO - __main__ - Step 137261: {'lr': 9.086084811093326e-06, 'samples': 26354112, 'steps': 137260, 'loss/train': 1.4323670864105225}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:09 - INFO - __main__ - Step 137265: {'lr': 9.080414946340071e-06, 'samples': 26354880, 'steps': 137264, 'loss/train': 1.5062495470046997}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:09 - INFO - __main__ - Step 137265: {'lr': 9.080414946340071e-06, 'samples': 26354880, 'steps': 137264, 'loss/train': 1.5062495470046997}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:12 - INFO - __main__ - Step 137272: {'lr': 9.070496862400468e-06, 'samples': 26356224, 'steps': 137271, 'loss/train': 0.984302818775177}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:14 - INFO - __main__ - Step 137277: {'lr': 9.063415773490014e-06, 'samples': 26357184, 'steps': 137276, 'loss/train': 1.2868765592575073}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:16 - INFO - __main__ - Step 137281: {'lr': 9.057752856479362e-06, 'samples': 26357952, 'steps': 137280, 'loss/train': 1.5077892541885376}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:16 - INFO - __main__ - Step 137281: {'lr': 9.057752856479362e-06, 'samples': 26357952, 'steps': 137280, 'loss/train': 1.5077892541885376}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:20 - INFO - __main__ - Step 137288: {'lr': 9.047846931482306e-06, 'samples': 26359296, 'steps': 137287, 'loss/train': 1.4085981845855713}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:22 - INFO - __main__ - Step 137292: {'lr': 9.042188791423628e-06, 'samples': 26360064, 'steps': 137291, 'loss/train': 0.8101268410682678}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:24 - INFO - __main__ - Step 137297: {'lr': 9.03511855922215e-06, 'samples': 26361024, 'steps': 137296, 'loss/train': 1.0777068138122559}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:24 - INFO - __main__ - Step 137297: {'lr': 9.03511855922215e-06, 'samples': 26361024, 'steps': 137296, 'loss/train': 1.0777068138122559}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:28 - INFO - __main__ - Step 137305: {'lr': 9.023811833635903e-06, 'samples': 26362560, 'steps': 137304, 'loss/train': 1.162799596786499}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:30 - INFO - __main__ - Step 137309: {'lr': 9.018161076746023e-06, 'samples': 26363328, 'steps': 137308, 'loss/train': 0.8667387366294861}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:32 - INFO - __main__ - Step 137313: {'lr': 9.012512057179345e-06, 'samples': 26364096, 'steps': 137312, 'loss/train': 1.289856195449829}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:32 - INFO - __main__ - Step 137313: {'lr': 9.012512057179345e-06, 'samples': 26364096, 'steps': 137312, 'loss/train': 1.289856195449829}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:36 - INFO - __main__ - Step 137321: {'lr': 9.001219230178354e-06, 'samples': 26365632, 'steps': 137320, 'loss/train': 1.452185869216919}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:38 - INFO - __main__ - Step 137325: {'lr': 8.995575422825448e-06, 'samples': 26366400, 'steps': 137324, 'loss/train': 1.6882961988449097}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:40 - INFO - __main__ - Step 137329: {'lr': 8.989933352958557e-06, 'samples': 26367168, 'steps': 137328, 'loss/train': 1.5628434419631958}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:42 - INFO - __main__ - Step 137334: {'lr': 8.982883209026598e-06, 'samples': 26368128, 'steps': 137333, 'loss/train': 1.4277551174163818}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:44 - INFO - __main__ - Step 137338: {'lr': 8.977245048651911e-06, 'samples': 26368896, 'steps': 137337, 'loss/train': 1.0988236665725708}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:47 - INFO - __main__ - Step 137342: {'lr': 8.971608625895356e-06, 'samples': 26369664, 'steps': 137341, 'loss/train': 1.3153977394104004}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:49 - INFO - __main__ - Step 137346: {'lr': 8.965973940797595e-06, 'samples': 26370432, 'steps': 137345, 'loss/train': 1.2228261232376099}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:50 - INFO - __main__ - Step 137350: {'lr': 8.960340993399208e-06, 'samples': 26371200, 'steps': 137349, 'loss/train': 1.0473403930664062}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:52 - INFO - __main__ - Step 137354: {'lr': 8.954709783740855e-06, 'samples': 26371968, 'steps': 137353, 'loss/train': 1.0431160926818848}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:55 - INFO - __main__ - Step 137359: {'lr': 8.947673215426666e-06, 'samples': 26372928, 'steps': 137358, 'loss/train': 1.3925299644470215}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:55 - INFO - __main__ - Step 137359: {'lr': 8.947673215426666e-06, 'samples': 26372928, 'steps': 137358, 'loss/train': 1.3925299644470215}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:35:58 - INFO - __main__ - Step 137366: {'lr': 8.937826581611769e-06, 'samples': 26374272, 'steps': 137365, 'loss/train': 0.3458479046821594}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:00 - INFO - __main__ - Step 137370: {'lr': 8.932202323319343e-06, 'samples': 26375040, 'steps': 137369, 'loss/train': 1.026477336883545}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:02 - INFO - __main__ - Step 137375: {'lr': 8.925174444440687e-06, 'samples': 26376000, 'steps': 137374, 'loss/train': 0.7156855463981628}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:05 - INFO - __main__ - Step 137380: {'lr': 8.91814928117693e-06, 'samples': 26376960, 'steps': 137379, 'loss/train': 1.0537664890289307}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:05 - INFO - __main__ - Step 137380: {'lr': 8.91814928117693e-06, 'samples': 26376960, 'steps': 137379, 'loss/train': 1.0537664890289307}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:09 - INFO - __main__ - Step 137387: {'lr': 8.908318614991417e-06, 'samples': 26378304, 'steps': 137386, 'loss/train': 0.7619863748550415}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:10 - INFO - __main__ - Step 137391: {'lr': 8.90270348135111e-06, 'samples': 26379072, 'steps': 137390, 'loss/train': 0.06907250732183456}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:12 - INFO - __main__ - Step 137395: {'lr': 8.897090085866338e-06, 'samples': 26379840, 'steps': 137394, 'loss/train': 1.3417702913284302}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:15 - INFO - __main__ - Step 137400: {'lr': 8.890075785853297e-06, 'samples': 26380800, 'steps': 137399, 'loss/train': 1.3897935152053833}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:17 - INFO - __main__ - Step 137404: {'lr': 8.88446630136644e-06, 'samples': 26381568, 'steps': 137403, 'loss/train': 0.5344817638397217}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:17 - INFO - __main__ - Step 137404: {'lr': 8.88446630136644e-06, 'samples': 26381568, 'steps': 137403, 'loss/train': 0.5344817638397217}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:20 - INFO - __main__ - Step 137411: {'lr': 8.874653886291967e-06, 'samples': 26382912, 'steps': 137410, 'loss/train': 0.7723680138587952}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:22 - INFO - __main__ - Step 137416: {'lr': 8.867648277789769e-06, 'samples': 26383872, 'steps': 137415, 'loss/train': 1.303730845451355}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:24 - INFO - __main__ - Step 137420: {'lr': 8.862045746693498e-06, 'samples': 26384640, 'steps': 137419, 'loss/train': 1.1257506608963013}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:27 - INFO - __main__ - Step 137425: {'lr': 8.85504502752138e-06, 'samples': 26385600, 'steps': 137424, 'loss/train': 1.3413821458816528}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:29 - INFO - __main__ - Step 137429: {'lr': 8.849446407991401e-06, 'samples': 26386368, 'steps': 137428, 'loss/train': 0.6657443046569824}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:29 - INFO - __main__ - Step 137429: {'lr': 8.849446407991401e-06, 'samples': 26386368, 'steps': 137428, 'loss/train': 0.6657443046569824}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:32 - INFO - __main__ - Step 137436: {'lr': 8.839653007198856e-06, 'samples': 26387712, 'steps': 137435, 'loss/train': 1.6222814321517944}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:35 - INFO - __main__ - Step 137441: {'lr': 8.832660980799795e-06, 'samples': 26388672, 'steps': 137440, 'loss/train': 1.7696504592895508}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:37 - INFO - __main__ - Step 137446: {'lr': 8.825671671057773e-06, 'samples': 26389632, 'steps': 137445, 'loss/train': 0.7415681481361389}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:37 - INFO - __main__ - Step 137446: {'lr': 8.825671671057773e-06, 'samples': 26389632, 'steps': 137445, 'loss/train': 0.7415681481361389}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:41 - INFO - __main__ - Step 137453: {'lr': 8.815891201552655e-06, 'samples': 26390976, 'steps': 137452, 'loss/train': 1.196619987487793}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:43 - INFO - __main__ - Step 137457: {'lr': 8.81030475264602e-06, 'samples': 26391744, 'steps': 137456, 'loss/train': 1.1707074642181396}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:45 - INFO - __main__ - Step 137461: {'lr': 8.804720042561082e-06, 'samples': 26392512, 'steps': 137460, 'loss/train': 0.3761148154735565}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:47 - INFO - __main__ - Step 137465: {'lr': 8.799137071338087e-06, 'samples': 26393280, 'steps': 137464, 'loss/train': 0.8637139797210693}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:49 - INFO - __main__ - Step 137469: {'lr': 8.793555839017253e-06, 'samples': 26394048, 'steps': 137468, 'loss/train': 1.5311049222946167}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:51 - INFO - __main__ - Step 137473: {'lr': 8.787976345638827e-06, 'samples': 26394816, 'steps': 137472, 'loss/train': 1.132435917854309}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:52 - INFO - __main__ - Step 137477: {'lr': 8.782398591243079e-06, 'samples': 26395584, 'steps': 137476, 'loss/train': 0.5512829422950745}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:55 - INFO - __main__ - Step 137482: {'lr': 8.775428843754041e-06, 'samples': 26396544, 'steps': 137481, 'loss/train': 1.2959511280059814}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:57 - INFO - __main__ - Step 137487: {'lr': 8.768461813566692e-06, 'samples': 26397504, 'steps': 137486, 'loss/train': 1.3925639390945435}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:36:57 - INFO - __main__ - Step 137487: {'lr': 8.768461813566692e-06, 'samples': 26397504, 'steps': 137486, 'loss/train': 1.3925639390945435}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:01 - INFO - __main__ - Step 137494: {'lr': 8.758712536520746e-06, 'samples': 26398848, 'steps': 137493, 'loss/train': 1.4074680805206299}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:02 - INFO - __main__ - Step 137498: {'lr': 8.753143912443363e-06, 'samples': 26399616, 'steps': 137497, 'loss/train': 1.7630926370620728}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:05 - INFO - __main__ - Step 137502: {'lr': 8.747577027599735e-06, 'samples': 26400384, 'steps': 137501, 'loss/train': 0.05986207723617554}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:07 - INFO - __main__ - Step 137507: {'lr': 8.740620867403848e-06, 'samples': 26401344, 'steps': 137506, 'loss/train': 1.0186762809753418}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:09 - INFO - __main__ - Step 137511: {'lr': 8.735057895982918e-06, 'samples': 26402112, 'steps': 137510, 'loss/train': 1.3262628316879272}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:11 - INFO - __main__ - Step 137515: {'lr': 8.72949666392614e-06, 'samples': 26402880, 'steps': 137514, 'loss/train': 1.2898688316345215}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:13 - INFO - __main__ - Step 137519: {'lr': 8.723937171273649e-06, 'samples': 26403648, 'steps': 137518, 'loss/train': 1.274928092956543}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:15 - INFO - __main__ - Step 137523: {'lr': 8.71837941806547e-06, 'samples': 26404416, 'steps': 137522, 'loss/train': 1.2954463958740234}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:17 - INFO - __main__ - Step 137528: {'lr': 8.711434672709878e-06, 'samples': 26405376, 'steps': 137527, 'loss/train': 0.5346939563751221}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:19 - INFO - __main__ - Step 137532: {'lr': 8.705880833397934e-06, 'samples': 26406144, 'steps': 137531, 'loss/train': 1.5472538471221924}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:21 - INFO - __main__ - Step 137536: {'lr': 8.700328733660473e-06, 'samples': 26406912, 'steps': 137535, 'loss/train': 1.371339201927185}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:23 - INFO - __main__ - Step 137540: {'lr': 8.694778373537576e-06, 'samples': 26407680, 'steps': 137539, 'loss/train': 1.2746615409851074}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:25 - INFO - __main__ - Step 137544: {'lr': 8.689229753069183e-06, 'samples': 26408448, 'steps': 137543, 'loss/train': 1.3772873878479004}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:27 - INFO - __main__ - Step 137549: {'lr': 8.682296423933894e-06, 'samples': 26409408, 'steps': 137548, 'loss/train': 0.9302381277084351}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:27 - INFO - __main__ - Step 137549: {'lr': 8.682296423933894e-06, 'samples': 26409408, 'steps': 137548, 'loss/train': 0.9302381277084351}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:31 - INFO - __main__ - Step 137556: {'lr': 8.672594329991345e-06, 'samples': 26410752, 'steps': 137555, 'loss/train': 0.755914032459259}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:33 - INFO - __main__ - Step 137560: {'lr': 8.667052668541092e-06, 'samples': 26411520, 'steps': 137559, 'loss/train': 1.140468955039978}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:35 - INFO - __main__ - Step 137565: {'lr': 8.660128038403187e-06, 'samples': 26412480, 'steps': 137564, 'loss/train': 1.2563596963882446}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:35 - INFO - __main__ - Step 137565: {'lr': 8.660128038403187e-06, 'samples': 26412480, 'steps': 137564, 'loss/train': 1.2563596963882446}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:38 - INFO - __main__ - Step 137572: {'lr': 8.650438123476573e-06, 'samples': 26413824, 'steps': 137571, 'loss/train': 1.371561050415039}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:41 - INFO - __main__ - Step 137576: {'lr': 8.644903421683587e-06, 'samples': 26414592, 'steps': 137575, 'loss/train': 1.4100788831710815}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:43 - INFO - __main__ - Step 137581: {'lr': 8.637987491341897e-06, 'samples': 26415552, 'steps': 137580, 'loss/train': 1.5259320735931396}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:45 - INFO - __main__ - Step 137585: {'lr': 8.632456704636804e-06, 'samples': 26416320, 'steps': 137584, 'loss/train': 1.280288815498352}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:47 - INFO - __main__ - Step 137589: {'lr': 8.626927658035522e-06, 'samples': 26417088, 'steps': 137588, 'loss/train': 1.3224495649337769}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:49 - INFO - __main__ - Step 137593: {'lr': 8.621400351577962e-06, 'samples': 26417856, 'steps': 137592, 'loss/train': 0.17687202990055084}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:51 - INFO - __main__ - Step 137597: {'lr': 8.61587478530393e-06, 'samples': 26418624, 'steps': 137596, 'loss/train': 0.8513680696487427}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:53 - INFO - __main__ - Step 137602: {'lr': 8.608970274655264e-06, 'samples': 26419584, 'steps': 137601, 'loss/train': 1.5999653339385986}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:55 - INFO - __main__ - Step 137606: {'lr': 8.603448623939857e-06, 'samples': 26420352, 'steps': 137605, 'loss/train': 1.4791717529296875}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:57 - INFO - __main__ - Step 137610: {'lr': 8.597928713537456e-06, 'samples': 26421120, 'steps': 137609, 'loss/train': 0.6518523097038269}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:37:59 - INFO - __main__ - Step 137614: {'lr': 8.592410543487833e-06, 'samples': 26421888, 'steps': 137613, 'loss/train': 1.0473413467407227}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:01 - INFO - __main__ - Step 137618: {'lr': 8.586894113830761e-06, 'samples': 26422656, 'steps': 137617, 'loss/train': 1.2463858127593994}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:03 - INFO - __main__ - Step 137623: {'lr': 8.580001024247025e-06, 'samples': 26423616, 'steps': 137622, 'loss/train': 1.2983877658843994}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:06 - INFO - __main__ - Step 137627: {'lr': 8.574488510618594e-06, 'samples': 26424384, 'steps': 137626, 'loss/train': 1.5294922590255737}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:06 - INFO - __main__ - Step 137627: {'lr': 8.574488510618594e-06, 'samples': 26424384, 'steps': 137626, 'loss/train': 1.5294922590255737}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:09 - INFO - __main__ - Step 137634: {'lr': 8.564845799923199e-06, 'samples': 26425728, 'steps': 137633, 'loss/train': 1.0786693096160889}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:11 - INFO - __main__ - Step 137638: {'lr': 8.559338072825162e-06, 'samples': 26426496, 'steps': 137637, 'loss/train': 1.0271508693695068}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:11 - INFO - __main__ - Step 137638: {'lr': 8.559338072825162e-06, 'samples': 26426496, 'steps': 137637, 'loss/train': 1.0271508693695068}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:15 - INFO - __main__ - Step 137647: {'lr': 8.54695205109718e-06, 'samples': 26428224, 'steps': 137646, 'loss/train': 0.8355200290679932}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:15 - INFO - __main__ - Step 137647: {'lr': 8.54695205109718e-06, 'samples': 26428224, 'steps': 137646, 'loss/train': 0.8355200290679932}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:19 - INFO - __main__ - Step 137654: {'lr': 8.537324571139898e-06, 'samples': 26429568, 'steps': 137653, 'loss/train': 0.9163202047348022}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:22 - INFO - __main__ - Step 137659: {'lr': 8.530451063710088e-06, 'samples': 26430528, 'steps': 137658, 'loss/train': 0.8417865037918091}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:22 - INFO - __main__ - Step 137659: {'lr': 8.530451063710088e-06, 'samples': 26430528, 'steps': 137658, 'loss/train': 0.8417865037918091}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:25 - INFO - __main__ - Step 137665: {'lr': 8.522206445282038e-06, 'samples': 26431680, 'steps': 137664, 'loss/train': 0.913759708404541}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:28 - INFO - __main__ - Step 137670: {'lr': 8.515338922089333e-06, 'samples': 26432640, 'steps': 137669, 'loss/train': 0.99848473072052}}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:30 - INFO - __main__ - Step 137674: {'lr': 8.50984686209813e-06, 'samples': 26433408, 'steps': 137673, 'loss/train': 0.5483201146125793}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:30 - INFO - __main__ - Step 137674: {'lr': 8.50984686209813e-06, 'samples': 26433408, 'steps': 137673, 'loss/train': 0.5483201146125793}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:33 - INFO - __main__ - Step 137681: {'lr': 8.50023994638921e-06, 'samples': 26434752, 'steps': 137680, 'loss/train': 1.3856513500213623}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:36 - INFO - __main__ - Step 137686: {'lr': 8.493381128209548e-06, 'samples': 26435712, 'steps': 137685, 'loss/train': 1.5618540048599243}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:38 - INFO - __main__ - Step 137690: {'lr': 8.487896032406933e-06, 'samples': 26436480, 'steps': 137689, 'loss/train': 1.5397181510925293}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:38 - INFO - __main__ - Step 137690: {'lr': 8.487896032406933e-06, 'samples': 26436480, 'steps': 137689, 'loss/train': 1.5397181510925293}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:41 - INFO - __main__ - Step 137697: {'lr': 8.47830130440877e-06, 'samples': 26437824, 'steps': 137696, 'loss/train': 1.3141921758651733}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:43 - INFO - __main__ - Step 137701: {'lr': 8.47282099685287e-06, 'samples': 26438592, 'steps': 137700, 'loss/train': 1.4076263904571533}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:45 - INFO - __main__ - Step 137706: {'lr': 8.465973061052067e-06, 'samples': 26439552, 'steps': 137705, 'loss/train': 0.6544806957244873}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:48 - INFO - __main__ - Step 137710: {'lr': 8.460496671374828e-06, 'samples': 26440320, 'steps': 137709, 'loss/train': 1.172226071357727}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:48 - INFO - __main__ - Step 137710: {'lr': 8.460496671374828e-06, 'samples': 26440320, 'steps': 137709, 'loss/train': 1.172226071357727}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:51 - INFO - __main__ - Step 137717: {'lr': 8.450917179571306e-06, 'samples': 26441664, 'steps': 137716, 'loss/train': 1.0965403318405151}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:53 - INFO - __main__ - Step 137722: {'lr': 8.444077950562318e-06, 'samples': 26442624, 'steps': 137721, 'loss/train': 1.3410998582839966}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:53 - INFO - __main__ - Step 137722: {'lr': 8.444077950562318e-06, 'samples': 26442624, 'steps': 137721, 'loss/train': 1.3410998582839966}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:57 - INFO - __main__ - Step 137729: {'lr': 8.434507601304542e-06, 'samples': 26443968, 'steps': 137728, 'loss/train': 0.8925405740737915}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:38:59 - INFO - __main__ - Step 137733: {'lr': 8.429041224891604e-06, 'samples': 26444736, 'steps': 137732, 'loss/train': 0.8685097098350525}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:02 - INFO - __main__ - Step 137738: {'lr': 8.422210703463302e-06, 'samples': 26445696, 'steps': 137737, 'loss/train': 1.0911765098571777}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:04 - INFO - __main__ - Step 137742: {'lr': 8.416748245638984e-06, 'samples': 26446464, 'steps': 137741, 'loss/train': 0.6703207492828369}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:04 - INFO - __main__ - Step 137742: {'lr': 8.416748245638984e-06, 'samples': 26446464, 'steps': 137741, 'loss/train': 0.6703207492828369}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:07 - INFO - __main__ - Step 137748: {'lr': 8.408557824525432e-06, 'samples': 26447616, 'steps': 137747, 'loss/train': 0.1379656046628952}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:10 - INFO - __main__ - Step 137753: {'lr': 8.40173546716963e-06, 'samples': 26448576, 'steps': 137752, 'loss/train': 0.18296682834625244}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:10 - INFO - __main__ - Step 137753: {'lr': 8.40173546716963e-06, 'samples': 26448576, 'steps': 137752, 'loss/train': 0.18296682834625244}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:14 - INFO - __main__ - Step 137760: {'lr': 8.39218873902764e-06, 'samples': 26449920, 'steps': 137759, 'loss/train': 1.1725614070892334}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:15 - INFO - __main__ - Step 137764: {'lr': 8.386735860814649e-06, 'samples': 26450688, 'steps': 137763, 'loss/train': 0.8264093995094299}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:17 - INFO - __main__ - Step 137768: {'lr': 8.381284724476995e-06, 'samples': 26451456, 'steps': 137767, 'loss/train': 1.551268458366394}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:17 - INFO - __main__ - Step 137768: {'lr': 8.381284724476995e-06, 'samples': 26451456, 'steps': 137767, 'loss/train': 1.551268458366394}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:21 - INFO - __main__ - Step 137776: {'lr': 8.370387677584912e-06, 'samples': 26452992, 'steps': 137775, 'loss/train': 1.0358268022537231}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:23 - INFO - __main__ - Step 137780: {'lr': 8.364941767109003e-06, 'samples': 26453760, 'steps': 137779, 'loss/train': 1.3217322826385498}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:25 - INFO - __main__ - Step 137784: {'lr': 8.359497598665555e-06, 'samples': 26454528, 'steps': 137783, 'loss/train': 1.2953416109085083}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:27 - INFO - __main__ - Step 137789: {'lr': 8.352694837904184e-06, 'samples': 26455488, 'steps': 137788, 'loss/train': 1.2511568069458008}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:30 - INFO - __main__ - Step 137793: {'lr': 8.347254589177234e-06, 'samples': 26456256, 'steps': 137792, 'loss/train': 1.1780179738998413}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:30 - INFO - __main__ - Step 137793: {'lr': 8.347254589177234e-06, 'samples': 26456256, 'steps': 137792, 'loss/train': 1.1780179738998413}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:33 - INFO - __main__ - Step 137800: {'lr': 8.337738346001128e-06, 'samples': 26457600, 'steps': 137799, 'loss/train': 1.3552974462509155}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:35 - INFO - __main__ - Step 137804: {'lr': 8.332302888308441e-06, 'samples': 26458368, 'steps': 137803, 'loss/train': 1.5611279010772705}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:35 - INFO - __main__ - Step 137804: {'lr': 8.332302888308441e-06, 'samples': 26458368, 'steps': 137803, 'loss/train': 1.5611279010772705}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:40 - INFO - __main__ - Step 137813: {'lr': 8.320079478726123e-06, 'samples': 26460096, 'steps': 137812, 'loss/train': 1.8061062097549438}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:42 - INFO - __main__ - Step 137817: {'lr': 8.314649683546893e-06, 'samples': 26460864, 'steps': 137816, 'loss/train': 1.2755168676376343}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:43 - INFO - __main__ - Step 137821: {'lr': 8.30922163076267e-06, 'samples': 26461632, 'steps': 137820, 'loss/train': 0.8120062351226807}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:45 - INFO - __main__ - Step 137825: {'lr': 8.303795320412616e-06, 'samples': 26462400, 'steps': 137824, 'loss/train': 1.6131477355957031}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:48 - INFO - __main__ - Step 137830: {'lr': 8.297014882832687e-06, 'samples': 26463360, 'steps': 137829, 'loss/train': 1.009566068649292}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:50 - INFO - __main__ - Step 137834: {'lr': 8.291592493102517e-06, 'samples': 26464128, 'steps': 137833, 'loss/train': 1.2454783916473389}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:52 - INFO - __main__ - Step 137838: {'lr': 8.286171845933583e-06, 'samples': 26464896, 'steps': 137837, 'loss/train': 1.6875569820404053}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:53 - INFO - __main__ - Step 137842: {'lr': 8.280752941365044e-06, 'samples': 26465664, 'steps': 137841, 'loss/train': 1.0420691967010498}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:55 - INFO - __main__ - Step 137846: {'lr': 8.275335779435845e-06, 'samples': 26466432, 'steps': 137845, 'loss/train': 1.0632084608078003}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:58 - INFO - __main__ - Step 137851: {'lr': 8.268566777670516e-06, 'samples': 26467392, 'steps': 137850, 'loss/train': 1.4199576377868652}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:39:58 - INFO - __main__ - Step 137851: {'lr': 8.268566777670516e-06, 'samples': 26467392, 'steps': 137850, 'loss/train': 1.4199576377868652}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:02 - INFO - __main__ - Step 137859: {'lr': 8.257742038741327e-06, 'samples': 26468928, 'steps': 137858, 'loss/train': 0.8585596084594727}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:04 - INFO - __main__ - Step 137863: {'lr': 8.252332283465142e-06, 'samples': 26469696, 'steps': 137862, 'loss/train': 1.6125977039337158}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:06 - INFO - __main__ - Step 137867: {'lr': 8.246924271033268e-06, 'samples': 26470464, 'steps': 137866, 'loss/train': 1.3749207258224487}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:08 - INFO - __main__ - Step 137871: {'lr': 8.241518001484705e-06, 'samples': 26471232, 'steps': 137870, 'loss/train': 2.603693723678589}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:10 - INFO - __main__ - Step 137876: {'lr': 8.234762615537988e-06, 'samples': 26472192, 'steps': 137875, 'loss/train': 1.3565118312835693}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:10 - INFO - __main__ - Step 137876: {'lr': 8.234762615537988e-06, 'samples': 26472192, 'steps': 137875, 'loss/train': 1.3565118312835693}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:14 - INFO - __main__ - Step 137883: {'lr': 8.225309650528395e-06, 'samples': 26473536, 'steps': 137882, 'loss/train': 1.1770747900009155}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:15 - INFO - __main__ - Step 137887: {'lr': 8.219910352902565e-06, 'samples': 26474304, 'steps': 137886, 'loss/train': 1.2350525856018066}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:18 - INFO - __main__ - Step 137892: {'lr': 8.213163682078361e-06, 'samples': 26475264, 'steps': 137891, 'loss/train': 1.8969799280166626}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:18 - INFO - __main__ - Step 137892: {'lr': 8.213163682078361e-06, 'samples': 26475264, 'steps': 137891, 'loss/train': 1.8969799280166626}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:18 - INFO - __main__ - Step 137892: {'lr': 8.213163682078361e-06, 'samples': 26475264, 'steps': 137891, 'loss/train': 1.8969799280166626}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:24 - INFO - __main__ - Step 137903: {'lr': 8.198330593568808e-06, 'samples': 26477376, 'steps': 137902, 'loss/train': 1.2930378913879395}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:26 - INFO - __main__ - Step 137907: {'lr': 8.192940011722189e-06, 'samples': 26478144, 'steps': 137906, 'loss/train': 1.0945591926574707}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:28 - INFO - __main__ - Step 137912: {'lr': 8.186204235895417e-06, 'samples': 26479104, 'steps': 137911, 'loss/train': 1.2214844226837158}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:30 - INFO - __main__ - Step 137916: {'lr': 8.18081757646652e-06, 'samples': 26479872, 'steps': 137915, 'loss/train': 0.9490886926651001}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:32 - INFO - __main__ - Step 137920: {'lr': 8.175432660397496e-06, 'samples': 26480640, 'steps': 137919, 'loss/train': 1.372708797454834}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:34 - INFO - __main__ - Step 137924: {'lr': 8.170049487727177e-06, 'samples': 26481408, 'steps': 137923, 'loss/train': 1.4620877504348755}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:36 - INFO - __main__ - Step 137928: {'lr': 8.164668058494334e-06, 'samples': 26482176, 'steps': 137927, 'loss/train': 1.6181063652038574}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:36 - INFO - __main__ - Step 137928: {'lr': 8.164668058494334e-06, 'samples': 26482176, 'steps': 137927, 'loss/train': 1.6181063652038574}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:36 - INFO - __main__ - Step 137928: {'lr': 8.164668058494334e-06, 'samples': 26482176, 'steps': 137927, 'loss/train': 1.6181063652038574}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:42 - INFO - __main__ - Step 137939: {'lr': 8.149878118020371e-06, 'samples': 26484288, 'steps': 137938, 'loss/train': 1.094005823135376}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:44 - INFO - __main__ - Step 137943: {'lr': 8.14450322702362e-06, 'samples': 26485056, 'steps': 137942, 'loss/train': 1.2760436534881592}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:46 - INFO - __main__ - Step 137948: {'lr': 8.137787065250202e-06, 'samples': 26486016, 'steps': 137947, 'loss/train': 1.4829459190368652}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:49 - INFO - __main__ - Step 137952: {'lr': 8.132416097456686e-06, 'samples': 26486784, 'steps': 137951, 'loss/train': 1.0144259929656982}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:49 - INFO - __main__ - Step 137952: {'lr': 8.132416097456686e-06, 'samples': 26486784, 'steps': 137951, 'loss/train': 1.0144259929656982}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:52 - INFO - __main__ - Step 137959: {'lr': 8.123021099640637e-06, 'samples': 26488128, 'steps': 137958, 'loss/train': 1.2632824182510376}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:55 - INFO - __main__ - Step 137964: {'lr': 8.116313656483825e-06, 'samples': 26489088, 'steps': 137963, 'loss/train': 1.7544571161270142}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:57 - INFO - __main__ - Step 137968: {'lr': 8.110949663757777e-06, 'samples': 26489856, 'steps': 137967, 'loss/train': 0.8715443015098572}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:40:57 - INFO - __main__ - Step 137968: {'lr': 8.110949663757777e-06, 'samples': 26489856, 'steps': 137967, 'loss/train': 0.8715443015098572}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:00 - INFO - __main__ - Step 137975: {'lr': 8.101566872682181e-06, 'samples': 26491200, 'steps': 137974, 'loss/train': 0.9927247762680054}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:02 - INFO - __main__ - Step 137980: {'lr': 8.09486814891569e-06, 'samples': 26492160, 'steps': 137979, 'loss/train': 1.0044583082199097}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:05 - INFO - __main__ - Step 137984: {'lr': 8.089511131875838e-06, 'samples': 26492928, 'steps': 137983, 'loss/train': 1.3170430660247803}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:05 - INFO - __main__ - Step 137984: {'lr': 8.089511131875838e-06, 'samples': 26492928, 'steps': 137983, 'loss/train': 1.3170430660247803}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:08 - INFO - __main__ - Step 137990: {'lr': 8.081478876362126e-06, 'samples': 26494080, 'steps': 137989, 'loss/train': 1.093482494354248}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:11 - INFO - __main__ - Step 137995: {'lr': 8.074788327725873e-06, 'samples': 26495040, 'steps': 137994, 'loss/train': 0.6350134015083313}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:12 - INFO - __main__ - Step 137999: {'lr': 8.069437850953043e-06, 'samples': 26495808, 'steps': 137998, 'loss/train': 1.4823566675186157}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:14 - INFO - __main__ - Step 138003: {'lr': 8.064089118343082e-06, 'samples': 26496576, 'steps': 138002, 'loss/train': 1.0617595911026}57}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:17 - INFO - __main__ - Step 138008: {'lr': 8.05740565536836e-06, 'samples': 26497536, 'steps': 138007, 'loss/train': 1.8804606199264526}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:17 - INFO - __main__ - Step 138008: {'lr': 8.05740565536836e-06, 'samples': 26497536, 'steps': 138007, 'loss/train': 1.8804606199264526}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:20 - INFO - __main__ - Step 138015: {'lr': 8.048053385875803e-06, 'samples': 26498880, 'steps': 138014, 'loss/train': 1.3426413536071777}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:22 - INFO - __main__ - Step 138019: {'lr': 8.042711630302707e-06, 'samples': 26499648, 'steps': 138018, 'loss/train': 1.406929850578308}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:25 - INFO - __main__ - Step 138024: {'lr': 8.036036888840804e-06, 'samples': 26500608, 'steps': 138023, 'loss/train': 1.4123194217681885}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:25 - INFO - __main__ - Step 138024: {'lr': 8.036036888840804e-06, 'samples': 26500608, 'steps': 138023, 'loss/train': 1.4123194217681885}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:28 - INFO - __main__ - Step 138031: {'lr': 8.026696829870589e-06, 'samples': 26501952, 'steps': 138030, 'loss/train': 1.171614170074463}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:30 - INFO - __main__ - Step 138035: {'lr': 8.021362051950532e-06, 'samples': 26502720, 'steps': 138034, 'loss/train': 0.07080969214439392}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:33 - INFO - __main__ - Step 138040: {'lr': 8.014696032771356e-06, 'samples': 26503680, 'steps': 138039, 'loss/train': 1.3440443277359009}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:33 - INFO - __main__ - Step 138040: {'lr': 8.014696032771356e-06, 'samples': 26503680, 'steps': 138039, 'loss/train': 1.3440443277359009}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:36 - INFO - __main__ - Step 138046: {'lr': 8.006700407912848e-06, 'samples': 26504832, 'steps': 138045, 'loss/train': 0.12280923128128052}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:39 - INFO - __main__ - Step 138051: {'lr': 8.000040385749196e-06, 'samples': 26505792, 'steps': 138050, 'loss/train': 2.0079288482666016}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:41 - INFO - __main__ - Step 138056: {'lr': 7.99338308962172e-06, 'samples': 26506752, 'steps': 138055, 'loss/train': 1.2635117769241333}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:41 - INFO - __main__ - Step 138056: {'lr': 7.99338308962172e-06, 'samples': 26506752, 'steps': 138055, 'loss/train': 1.2635117769241333}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:45 - INFO - __main__ - Step 138063: {'lr': 7.984067454926802e-06, 'samples': 26508096, 'steps': 138062, 'loss/train': 1.1175386905670166}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:46 - INFO - __main__ - Step 138067: {'lr': 7.97874663415818e-06, 'samples': 26508864, 'steps': 138066, 'loss/train': 1.1728953123092651}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:48 - INFO - __main__ - Step 138071: {'lr': 7.97342755820618e-06, 'samples': 26509632, 'steps': 138070, 'loss/train': 1.1391229629516602}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:48 - INFO - __main__ - Step 138071: {'lr': 7.97342755820618e-06, 'samples': 26509632, 'steps': 138070, 'loss/train': 1.1391229629516602}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:48 - INFO - __main__ - Step 138071: {'lr': 7.97342755820618e-06, 'samples': 26509632, 'steps': 138070, 'loss/train': 1.1391229629516602}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:54 - INFO - __main__ - Step 138082: {'lr': 7.958809096362207e-06, 'samples': 26511744, 'steps': 138081, 'loss/train': 1.2418618202209473}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:57 - INFO - __main__ - Step 138087: {'lr': 7.952168703331708e-06, 'samples': 26512704, 'steps': 138086, 'loss/train': 1.3691320419311523}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:41:59 - INFO - __main__ - Step 138091: {'lr': 7.946858352038038e-06, 'samples': 26513472, 'steps': 138090, 'loss/train': 1.2585519552230835}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:01 - INFO - __main__ - Step 138095: {'lr': 7.941549745790922e-06, 'samples': 26514240, 'steps': 138094, 'loss/train': 1.4370598793029785}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:01 - INFO - __main__ - Step 138095: {'lr': 7.941549745790922e-06, 'samples': 26514240, 'steps': 138094, 'loss/train': 1.4370598793029785}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:04 - INFO - __main__ - Step 138102: {'lr': 7.93226388399193e-06, 'samples': 26515584, 'steps': 138101, 'loss/train': 1.2636311054229736}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:07 - INFO - __main__ - Step 138107: {'lr': 7.925634397711685e-06, 'samples': 26516544, 'steps': 138106, 'loss/train': 1.3167325258255005}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:09 - INFO - __main__ - Step 138112: {'lr': 7.91900763830583e-06, 'samples': 26517504, 'steps': 138111, 'loss/train': 1.2217977046966553}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:11 - INFO - __main__ - Step 138116: {'lr': 7.913708194180803e-06, 'samples': 26518272, 'steps': 138115, 'loss/train': 0.9049156904220581}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:11 - INFO - __main__ - Step 138116: {'lr': 7.913708194180803e-06, 'samples': 26518272, 'steps': 138115, 'loss/train': 0.9049156904220581}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:15 - INFO - __main__ - Step 138123: {'lr': 7.904438366578364e-06, 'samples': 26519616, 'steps': 138122, 'loss/train': 1.2175843715667725}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:17 - INFO - __main__ - Step 138127: {'lr': 7.899143722080532e-06, 'samples': 26520384, 'steps': 138126, 'loss/train': 0.7007185220718384}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:17 - INFO - __main__ - Step 138127: {'lr': 7.899143722080532e-06, 'samples': 26520384, 'steps': 138126, 'loss/train': 0.7007185220718384}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:20 - INFO - __main__ - Step 138135: {'lr': 7.888559669294664e-06, 'samples': 26521920, 'steps': 138134, 'loss/train': 1.201690435409546}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:22 - INFO - __main__ - Step 138139: {'lr': 7.883270261082987e-06, 'samples': 26522688, 'steps': 138138, 'loss/train': 0.7128851413726807}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:25 - INFO - __main__ - Step 138144: {'lr': 7.876660955439208e-06, 'samples': 26523648, 'steps': 138143, 'loss/train': 1.4134846925735474}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:27 - INFO - __main__ - Step 138148: {'lr': 7.871375474667347e-06, 'samples': 26524416, 'steps': 138147, 'loss/train': 1.5962231159210205}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:27 - INFO - __main__ - Step 138148: {'lr': 7.871375474667347e-06, 'samples': 26524416, 'steps': 138147, 'loss/train': 1.5962231159210205}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:30 - INFO - __main__ - Step 138155: {'lr': 7.862130083667263e-06, 'samples': 26525760, 'steps': 138154, 'loss/train': 0.9502419233322144}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:32 - INFO - __main__ - Step 138159: {'lr': 7.856849403361621e-06, 'samples': 26526528, 'steps': 138158, 'loss/train': 0.9483413100242615}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:35 - INFO - __main__ - Step 138164: {'lr': 7.85025100786832e-06, 'samples': 26527488, 'steps': 138163, 'loss/train': 1.1873611211776733}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:37 - INFO - __main__ - Step 138168: {'lr': 7.844974255430987e-06, 'samples': 26528256, 'steps': 138167, 'loss/train': 1.5493189096450806}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:37 - INFO - __main__ - Step 138168: {'lr': 7.844974255430987e-06, 'samples': 26528256, 'steps': 138167, 'loss/train': 1.5493189096450806}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:40 - INFO - __main__ - Step 138175: {'lr': 7.835744139474299e-06, 'samples': 26529600, 'steps': 138174, 'loss/train': 1.325795292854309}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:43 - INFO - __main__ - Step 138180: {'lr': 7.829154472957706e-06, 'samples': 26530560, 'steps': 138179, 'loss/train': 1.4218558073043823}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:43 - INFO - __main__ - Step 138180: {'lr': 7.829154472957706e-06, 'samples': 26530560, 'steps': 138179, 'loss/train': 1.4218558073043823}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:47 - INFO - __main__ - Step 138188: {'lr': 7.818616680721147e-06, 'samples': 26532096, 'steps': 138187, 'loss/train': 1.2183688879013062}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:49 - INFO - __main__ - Step 138192: {'lr': 7.813350403540559e-06, 'samples': 26532864, 'steps': 138191, 'loss/train': 1.2346395254135132}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:50 - INFO - __main__ - Step 138196: {'lr': 7.808085872369003e-06, 'samples': 26533632, 'steps': 138195, 'loss/train': 1.544091820716858}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:53 - INFO - __main__ - Step 138201: {'lr': 7.801507663787676e-06, 'samples': 26534592, 'steps': 138200, 'loss/train': 1.301986813545227}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:53 - INFO - __main__ - Step 138201: {'lr': 7.801507663787676e-06, 'samples': 26534592, 'steps': 138200, 'loss/train': 1.301986813545227}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:57 - INFO - __main__ - Step 138209: {'lr': 7.790988204895321e-06, 'samples': 26536128, 'steps': 138208, 'loss/train': 0.9488916993141174}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:42:57 - INFO - __main__ - Step 138209: {'lr': 7.790988204895321e-06, 'samples': 26536128, 'steps': 138208, 'loss/train': 0.9488916993141174}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:00 - INFO - __main__ - Step 138216: {'lr': 7.781789407975386e-06, 'samples': 26537472, 'steps': 138215, 'loss/train': 1.0483791828155518}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:03 - INFO - __main__ - Step 138221: {'lr': 7.775222112928692e-06, 'samples': 26538432, 'steps': 138220, 'loss/train': 1.2643342018127441}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:06 - INFO - __main__ - Step 138226: {'lr': 7.768657546450648e-06, 'samples': 26539392, 'steps': 138225, 'loss/train': 1.1169673204421997}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:08 - INFO - __main__ - Step 138230: {'lr': 7.763407857887344e-06, 'samples': 26540160, 'steps': 138229, 'loss/train': 1.1284592151641846}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:10 - INFO - __main__ - Step 138234: {'lr': 7.758159915693203e-06, 'samples': 26540928, 'steps': 138233, 'loss/train': 1.436363697052002}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:11 - INFO - __main__ - Step 138238: {'lr': 7.752913719905996e-06, 'samples': 26541696, 'steps': 138237, 'loss/train': 1.2333272695541382}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:11 - INFO - __main__ - Step 138238: {'lr': 7.752913719905996e-06, 'samples': 26541696, 'steps': 138237, 'loss/train': 1.2333272695541382}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:11 - INFO - __main__ - Step 138238: {'lr': 7.752913719905996e-06, 'samples': 26541696, 'steps': 138237, 'loss/train': 1.2333272695541382}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:17 - INFO - __main__ - Step 138248: {'lr': 7.739805871216616e-06, 'samples': 26543616, 'steps': 138247, 'loss/train': 1.4209973812103271}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:19 - INFO - __main__ - Step 138252: {'lr': 7.734565788151543e-06, 'samples': 26544384, 'steps': 138251, 'loss/train': 0.9684511423110962}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:21 - INFO - __main__ - Step 138257: {'lr': 7.728018140448629e-06, 'samples': 26545344, 'steps': 138256, 'loss/train': 1.1924747228622437}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:24 - INFO - __main__ - Step 138261: {'lr': 7.722781987235028e-06, 'samples': 26546112, 'steps': 138260, 'loss/train': 1.2350317239761353}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:25 - INFO - __main__ - Step 138265: {'lr': 7.717547580683437e-06, 'samples': 26546880, 'steps': 138264, 'loss/train': 0.5941764116287231}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:27 - INFO - __main__ - Step 138269: {'lr': 7.712314920831603e-06, 'samples': 26547648, 'steps': 138268, 'loss/train': 1.3485815525054932}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:29 - INFO - __main__ - Step 138273: {'lr': 7.707084007717274e-06, 'samples': 26548416, 'steps': 138272, 'loss/train': 1.7258669137954712}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:31 - INFO - __main__ - Step 138277: {'lr': 7.701854841378114e-06, 'samples': 26549184, 'steps': 138276, 'loss/train': 1.1643110513687134}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:33 - INFO - __main__ - Step 138281: {'lr': 7.696627421851816e-06, 'samples': 26549952, 'steps': 138280, 'loss/train': 1.2049559354782104}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:35 - INFO - __main__ - Step 138285: {'lr': 7.691401749176125e-06, 'samples': 26550720, 'steps': 138284, 'loss/train': 2.7029104232788086}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:37 - INFO - __main__ - Step 138290: {'lr': 7.68487211489749e-06, 'samples': 26551680, 'steps': 138289, 'loss/train': 0.9556211233139038}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:40 - INFO - __main__ - Step 138294: {'lr': 7.679650372773268e-06, 'samples': 26552448, 'steps': 138293, 'loss/train': 1.5747075080871582}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:42 - INFO - __main__ - Step 138298: {'lr': 7.674430377622054e-06, 'samples': 26553216, 'steps': 138297, 'loss/train': 1.0404694080352783}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:44 - INFO - __main__ - Step 138302: {'lr': 7.669212129481407e-06, 'samples': 26553984, 'steps': 138301, 'loss/train': 1.1826231479644775}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:45 - INFO - __main__ - Step 138306: {'lr': 7.663995628388986e-06, 'samples': 26554752, 'steps': 138305, 'loss/train': 1.1017532348632812}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:47 - INFO - __main__ - Step 138310: {'lr': 7.658780874382376e-06, 'samples': 26555520, 'steps': 138309, 'loss/train': 1.3637281656265259}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:50 - INFO - __main__ - Step 138315: {'lr': 7.652264888770854e-06, 'samples': 26556480, 'steps': 138314, 'loss/train': 0.709363579750061}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:50 - INFO - __main__ - Step 138315: {'lr': 7.652264888770854e-06, 'samples': 26556480, 'steps': 138314, 'loss/train': 0.709363579750061}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:53 - INFO - __main__ - Step 138322: {'lr': 7.643147095253434e-06, 'samples': 26557824, 'steps': 138321, 'loss/train': 1.3346140384674072}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:55 - INFO - __main__ - Step 138326: {'lr': 7.637939329965954e-06, 'samples': 26558592, 'steps': 138325, 'loss/train': 1.3752700090408325}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:57 - INFO - __main__ - Step 138331: {'lr': 7.631432080464602e-06, 'samples': 26559552, 'steps': 138330, 'loss/train': 1.0748618841171265}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:43:57 - INFO - __main__ - Step 138331: {'lr': 7.631432080464602e-06, 'samples': 26559552, 'steps': 138330, 'loss/train': 1.0748618841171265}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:02 - INFO - __main__ - Step 138339: {'lr': 7.6210261600848375e-06, 'samples': 26561088, 'steps': 138338, 'loss/train': 1.4355993270874023}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:03 - INFO - __main__ - Step 138343: {'lr': 7.615825820969618e-06, 'samples': 26561856, 'steps': 138342, 'loss/train': 1.373570442199707}3}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:05 - INFO - __main__ - Step 138347: {'lr': 7.610627229287514e-06, 'samples': 26562624, 'steps': 138346, 'loss/train': 0.849307656288147}3}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:05 - INFO - __main__ - Step 138347: {'lr': 7.610627229287514e-06, 'samples': 26562624, 'steps': 138346, 'loss/train': 0.849307656288147}3}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:10 - INFO - __main__ - Step 138355: {'lr': 7.600235288372448e-06, 'samples': 26564160, 'steps': 138354, 'loss/train': 1.2602914571762085}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:12 - INFO - __main__ - Step 138359: {'lr': 7.5950419392144254e-06, 'samples': 26564928, 'steps': 138358, 'loss/train': 0.8206194043159485}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:13 - INFO - __main__ - Step 138363: {'lr': 7.589850337639343e-06, 'samples': 26565696, 'steps': 138362, 'loss/train': 1.6639765501022339}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:15 - INFO - __main__ - Step 138367: {'lr': 7.584660483684669e-06, 'samples': 26566464, 'steps': 138366, 'loss/train': 1.354425072669983}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:18 - INFO - __main__ - Step 138372: {'lr': 7.578175623889405e-06, 'samples': 26567424, 'steps': 138371, 'loss/train': 1.1310211420059204}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:20 - INFO - __main__ - Step 138376: {'lr': 7.572989702217259e-06, 'samples': 26568192, 'steps': 138375, 'loss/train': 1.0480924844741821}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:22 - INFO - __main__ - Step 138380: {'lr': 7.567805528287092e-06, 'samples': 26568960, 'steps': 138379, 'loss/train': 0.914145290851593}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:22 - INFO - __main__ - Step 138380: {'lr': 7.567805528287092e-06, 'samples': 26568960, 'steps': 138379, 'loss/train': 0.914145290851593}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:25 - INFO - __main__ - Step 138387: {'lr': 7.558737429525647e-06, 'samples': 26570304, 'steps': 138386, 'loss/train': 1.1440626382827759}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:27 - INFO - __main__ - Step 138392: {'lr': 7.55226349332172e-06, 'samples': 26571264, 'steps': 138391, 'loss/train': 1.292283535003662}9}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:30 - INFO - __main__ - Step 138397: {'lr': 7.5457922881979616e-06, 'samples': 26572224, 'steps': 138396, 'loss/train': 1.6549967527389526}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:32 - INFO - __main__ - Step 138401: {'lr': 7.540617290525742e-06, 'samples': 26572992, 'steps': 138400, 'loss/train': 1.125745415687561}6}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:32 - INFO - __main__ - Step 138401: {'lr': 7.540617290525742e-06, 'samples': 26572992, 'steps': 138400, 'loss/train': 1.125745415687561}6}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:35 - INFO - __main__ - Step 138408: {'lr': 7.531565250687322e-06, 'samples': 26574336, 'steps': 138407, 'loss/train': 0.8955969214439392}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:38 - INFO - __main__ - Step 138413: {'lr': 7.525102785509996e-06, 'samples': 26575296, 'steps': 138412, 'loss/train': 1.1145192384719849}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:40 - INFO - __main__ - Step 138418: {'lr': 7.518643051718843e-06, 'samples': 26576256, 'steps': 138417, 'loss/train': 1.2166677713394165}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:40 - INFO - __main__ - Step 138418: {'lr': 7.518643051718843e-06, 'samples': 26576256, 'steps': 138417, 'loss/train': 1.2166677713394165}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:44 - INFO - __main__ - Step 138425: {'lr': 7.509604013278576e-06, 'samples': 26577600, 'steps': 138424, 'loss/train': 0.7540420293807983}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:45 - INFO - __main__ - Step 138429: {'lr': 7.504441252216221e-06, 'samples': 26578368, 'steps': 138428, 'loss/train': 1.378525972366333}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:47 - INFO - __main__ - Step 138433: {'lr': 7.499280239389977e-06, 'samples': 26579136, 'steps': 138432, 'loss/train': 1.3333487510681152}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:50 - INFO - __main__ - Step 138438: {'lr': 7.492831431870878e-06, 'samples': 26580096, 'steps': 138437, 'loss/train': 1.313600778579712}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:52 - INFO - __main__ - Step 138442: {'lr': 7.487674352711915e-06, 'samples': 26580864, 'steps': 138441, 'loss/train': 1.2581489086151123}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:54 - INFO - __main__ - Step 138446: {'lr': 7.482519021909939e-06, 'samples': 26581632, 'steps': 138445, 'loss/train': 1.4599180221557617}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:55 - INFO - __main__ - Step 138450: {'lr': 7.4773654395020875e-06, 'samples': 26582400, 'steps': 138449, 'loss/train': 1.2865173816680908}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:44:58 - INFO - __main__ - Step 138454: {'lr': 7.472213605525552e-06, 'samples': 26583168, 'steps': 138453, 'loss/train': 1.1633268594741821}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:00 - INFO - __main__ - Step 138459: {'lr': 7.465776271843028e-06, 'samples': 26584128, 'steps': 138458, 'loss/train': 1.316955327987671}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:00 - INFO - __main__ - Step 138459: {'lr': 7.465776271843028e-06, 'samples': 26584128, 'steps': 138458, 'loss/train': 1.316955327987671}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:04 - INFO - __main__ - Step 138466: {'lr': 7.456768594555158e-06, 'samples': 26585472, 'steps': 138465, 'loss/train': 0.8476134538650513}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:05 - INFO - __main__ - Step 138470: {'lr': 7.451623754675147e-06, 'samples': 26586240, 'steps': 138469, 'loss/train': 1.2324849367141724}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:07 - INFO - __main__ - Step 138474: {'lr': 7.446480663412053e-06, 'samples': 26587008, 'steps': 138473, 'loss/train': 1.167197823524475}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:07 - INFO - __main__ - Step 138474: {'lr': 7.446480663412053e-06, 'samples': 26587008, 'steps': 138473, 'loss/train': 1.167197823524475}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:07 - INFO - __main__ - Step 138474: {'lr': 7.446480663412053e-06, 'samples': 26587008, 'steps': 138473, 'loss/train': 1.167197823524475}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:13 - INFO - __main__ - Step 138485: {'lr': 7.432346179046923e-06, 'samples': 26589120, 'steps': 138484, 'loss/train': 1.2928920984268188}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:13 - INFO - __main__ - Step 138485: {'lr': 7.432346179046923e-06, 'samples': 26589120, 'steps': 138484, 'loss/train': 1.2928920984268188}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:18 - INFO - __main__ - Step 138493: {'lr': 7.422074860600509e-06, 'samples': 26590656, 'steps': 138492, 'loss/train': 1.4017754793167114}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:19 - INFO - __main__ - Step 138497: {'lr': 7.416941824603646e-06, 'samples': 26591424, 'steps': 138496, 'loss/train': 1.4058046340942383}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:21 - INFO - __main__ - Step 138501: {'lr': 7.411810537473751e-06, 'samples': 26592192, 'steps': 138500, 'loss/train': 1.721417784690857}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:21 - INFO - __main__ - Step 138501: {'lr': 7.411810537473751e-06, 'samples': 26592192, 'steps': 138500, 'loss/train': 1.721417784690857}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:26 - INFO - __main__ - Step 138510: {'lr': 7.4002715359176855e-06, 'samples': 26593920, 'steps': 138509, 'loss/train': 1.5250905752182007}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:28 - INFO - __main__ - Step 138514: {'lr': 7.395145932860669e-06, 'samples': 26594688, 'steps': 138513, 'loss/train': 0.9658278822898865}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:30 - INFO - __main__ - Step 138518: {'lr': 7.390022078827718e-06, 'samples': 26595456, 'steps': 138517, 'loss/train': 0.8816930651664734}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:32 - INFO - __main__ - Step 138522: {'lr': 7.384899973855746e-06, 'samples': 26596224, 'steps': 138521, 'loss/train': 1.5031061172485352}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:34 - INFO - __main__ - Step 138527: {'lr': 7.378499802314115e-06, 'samples': 26597184, 'steps': 138526, 'loss/train': 1.2988719940185547}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:34 - INFO - __main__ - Step 138527: {'lr': 7.378499802314115e-06, 'samples': 26597184, 'steps': 138526, 'loss/train': 1.2988719940185547}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:38 - INFO - __main__ - Step 138534: {'lr': 7.369544153675078e-06, 'samples': 26598528, 'steps': 138533, 'loss/train': 1.5309057235717773}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:40 - INFO - __main__ - Step 138538: {'lr': 7.364429045316256e-06, 'samples': 26599296, 'steps': 138537, 'loss/train': 1.5037864446640015}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:42 - INFO - __main__ - Step 138542: {'lr': 7.3593156862029605e-06, 'samples': 26600064, 'steps': 138541, 'loss/train': 0.6697014570236206}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:44 - INFO - __main__ - Step 138547: {'lr': 7.352926447243946e-06, 'samples': 26601024, 'steps': 138546, 'loss/train': 1.135678768157959}6}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:47 - INFO - __main__ - Step 138551: {'lr': 7.347817024067782e-06, 'samples': 26601792, 'steps': 138550, 'loss/train': 0.9740868806838989}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:47 - INFO - __main__ - Step 138551: {'lr': 7.347817024067782e-06, 'samples': 26601792, 'steps': 138550, 'loss/train': 0.9740868806838989}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:50 - INFO - __main__ - Step 138558: {'lr': 7.338879742941839e-06, 'samples': 26603136, 'steps': 138557, 'loss/train': 1.1648753881454468}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:52 - INFO - __main__ - Step 138562: {'lr': 7.333775130608733e-06, 'samples': 26603904, 'steps': 138561, 'loss/train': 1.3678518533706665}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:54 - INFO - __main__ - Step 138567: {'lr': 7.3273968253839695e-06, 'samples': 26604864, 'steps': 138566, 'loss/train': 1.1184624433517456}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:54 - INFO - __main__ - Step 138567: {'lr': 7.3273968253839695e-06, 'samples': 26604864, 'steps': 138566, 'loss/train': 1.1184624433517456}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:45:57 - INFO - __main__ - Step 138574: {'lr': 7.318471790555642e-06, 'samples': 26606208, 'steps': 138573, 'loss/train': 1.6570452451705933}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:00 - INFO - __main__ - Step 138578: {'lr': 7.3133741763092285e-06, 'samples': 26606976, 'steps': 138577, 'loss/train': 0.7364013195037842}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:02 - INFO - __main__ - Step 138583: {'lr': 7.30700461889966e-06, 'samples': 26607936, 'steps': 138582, 'loss/train': 1.5943248271942139}2}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:02 - INFO - __main__ - Step 138583: {'lr': 7.30700461889966e-06, 'samples': 26607936, 'steps': 138582, 'loss/train': 1.5943248271942139}2}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:07 - INFO - __main__ - Step 138591: {'lr': 7.296819013467515e-06, 'samples': 26609472, 'steps': 138590, 'loss/train': 1.4908733367919922}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:09 - INFO - __main__ - Step 138595: {'lr': 7.291728835332468e-06, 'samples': 26610240, 'steps': 138594, 'loss/train': 1.6905194520950317}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:10 - INFO - __main__ - Step 138599: {'lr': 7.286640406967054e-06, 'samples': 26611008, 'steps': 138598, 'loss/train': 1.6368967294692993}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:12 - INFO - __main__ - Step 138603: {'lr': 7.281553728407997e-06, 'samples': 26611776, 'steps': 138602, 'loss/train': 1.3202815055847168}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:15 - INFO - __main__ - Step 138608: {'lr': 7.275197840930087e-06, 'samples': 26612736, 'steps': 138607, 'loss/train': 1.2826405763626099}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:15 - INFO - __main__ - Step 138608: {'lr': 7.275197840930087e-06, 'samples': 26612736, 'steps': 138607, 'loss/train': 1.2826405763626099}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:18 - INFO - __main__ - Step 138615: {'lr': 7.266304191935219e-06, 'samples': 26614080, 'steps': 138614, 'loss/train': 0.8859825730323792}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:20 - INFO - __main__ - Step 138619: {'lr': 7.261224512967956e-06, 'samples': 26614848, 'steps': 138618, 'loss/train': 1.3212538957595825}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:22 - INFO - __main__ - Step 138623: {'lr': 7.256146583990264e-06, 'samples': 26615616, 'steps': 138622, 'loss/train': 0.8594983220100403}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:22 - INFO - __main__ - Step 138623: {'lr': 7.256146583990264e-06, 'samples': 26615616, 'steps': 138622, 'loss/train': 0.8594983220100403}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:26 - INFO - __main__ - Step 138630: {'lr': 7.24726441930168e-06, 'samples': 26616960, 'steps': 138629, 'loss/train': 1.6288269758224487}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:28 - INFO - __main__ - Step 138635: {'lr': 7.240923297360397e-06, 'samples': 26617920, 'steps': 138634, 'loss/train': 1.5515468120574951}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:28 - INFO - __main__ - Step 138635: {'lr': 7.240923297360397e-06, 'samples': 26617920, 'steps': 138634, 'loss/train': 1.5515468120574951}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:32 - INFO - __main__ - Step 138643: {'lr': 7.230783190225687e-06, 'samples': 26619456, 'steps': 138642, 'loss/train': 1.4805846214294434}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:34 - INFO - __main__ - Step 138647: {'lr': 7.225715761953605e-06, 'samples': 26620224, 'steps': 138646, 'loss/train': 1.4424717426300049}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:36 - INFO - __main__ - Step 138651: {'lr': 7.220650083927027e-06, 'samples': 26620992, 'steps': 138650, 'loss/train': 1.3632166385650635}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:38 - INFO - __main__ - Step 138655: {'lr': 7.215586156182508e-06, 'samples': 26621760, 'steps': 138654, 'loss/train': 1.2298911809921265}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:38 - INFO - __main__ - Step 138655: {'lr': 7.215586156182508e-06, 'samples': 26621760, 'steps': 138654, 'loss/train': 1.2298911809921265}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:42 - INFO - __main__ - Step 138663: {'lr': 7.205463551685665e-06, 'samples': 26623296, 'steps': 138662, 'loss/train': 1.6024528741836548}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:42 - INFO - __main__ - Step 138663: {'lr': 7.205463551685665e-06, 'samples': 26623296, 'steps': 138662, 'loss/train': 1.6024528741836548}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:45 - INFO - __main__ - Step 138670: {'lr': 7.196612016213189e-06, 'samples': 26624640, 'steps': 138669, 'loss/train': 1.5716921091079712}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:48 - INFO - __main__ - Step 138676: {'lr': 7.189029252535784e-06, 'samples': 26625792, 'steps': 138675, 'loss/train': 1.7336599826812744}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:48 - INFO - __main__ - Step 138676: {'lr': 7.189029252535784e-06, 'samples': 26625792, 'steps': 138675, 'loss/train': 1.7336599826812744}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:52 - INFO - __main__ - Step 138683: {'lr': 7.180187672933603e-06, 'samples': 26627136, 'steps': 138682, 'loss/train': 0.625313401222229}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:54 - INFO - __main__ - Step 138687: {'lr': 7.175137748758748e-06, 'samples': 26627904, 'steps': 138686, 'loss/train': 1.2997101545333862}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:56 - INFO - __main__ - Step 138692: {'lr': 7.1688278053400615e-06, 'samples': 26628864, 'steps': 138691, 'loss/train': 1.5109418630599976}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:46:56 - INFO - __main__ - Step 138692: {'lr': 7.1688278053400615e-06, 'samples': 26628864, 'steps': 138691, 'loss/train': 1.5109418630599976}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:00 - INFO - __main__ - Step 138699: {'lr': 7.159998480040897e-06, 'samples': 26630208, 'steps': 138698, 'loss/train': 1.1038174629211426}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:02 - INFO - __main__ - Step 138703: {'lr': 7.154955558525078e-06, 'samples': 26630976, 'steps': 138702, 'loss/train': 1.5253950357437134}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:05 - INFO - __main__ - Step 138707: {'lr': 7.149914387764938e-06, 'samples': 26631744, 'steps': 138706, 'loss/train': 1.1912058591842651}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:05 - INFO - __main__ - Step 138707: {'lr': 7.149914387764938e-06, 'samples': 26631744, 'steps': 138706, 'loss/train': 1.1912058591842651}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:08 - INFO - __main__ - Step 138714: {'lr': 7.141096551799814e-06, 'samples': 26633088, 'steps': 138713, 'loss/train': 0.28070810437202454}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:10 - INFO - __main__ - Step 138719: {'lr': 7.134801380381916e-06, 'samples': 26634048, 'steps': 138718, 'loss/train': 1.0094668865203857}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:10 - INFO - __main__ - Step 138719: {'lr': 7.134801380381916e-06, 'samples': 26634048, 'steps': 138718, 'loss/train': 1.0094668865203857}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:15 - INFO - __main__ - Step 138727: {'lr': 7.124734796570887e-06, 'samples': 26635584, 'steps': 138726, 'loss/train': 1.4723252058029175}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:17 - INFO - __main__ - Step 138731: {'lr': 7.119704131107602e-06, 'samples': 26636352, 'steps': 138730, 'loss/train': 1.2717233896255493}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:18 - INFO - __main__ - Step 138735: {'lr': 7.114675216654071e-06, 'samples': 26637120, 'steps': 138734, 'loss/train': 1.4675180912017822}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:20 - INFO - __main__ - Step 138739: {'lr': 7.109648053246626e-06, 'samples': 26637888, 'steps': 138738, 'loss/train': 1.1906448602676392}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:23 - INFO - __main__ - Step 138744: {'lr': 7.10336656145108e-06, 'samples': 26638848, 'steps': 138743, 'loss/train': 1.507412075996399}2}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:25 - INFO - __main__ - Step 138748: {'lr': 7.098343338029778e-06, 'samples': 26639616, 'steps': 138747, 'loss/train': 0.9358780384063721}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:25 - INFO - __main__ - Step 138748: {'lr': 7.098343338029778e-06, 'samples': 26639616, 'steps': 138747, 'loss/train': 0.9358780384063721}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:28 - INFO - __main__ - Step 138755: {'lr': 7.089556910802236e-06, 'samples': 26640960, 'steps': 138754, 'loss/train': 1.1851946115493774}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:31 - INFO - __main__ - Step 138760: {'lr': 7.083284174893567e-06, 'samples': 26641920, 'steps': 138759, 'loss/train': 1.3986599445343018}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:33 - INFO - __main__ - Step 138764: {'lr': 7.0782679563446694e-06, 'samples': 26642688, 'steps': 138763, 'loss/train': 0.8975769281387329}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:35 - INFO - __main__ - Step 138768: {'lr': 7.0732534891043424e-06, 'samples': 26643456, 'steps': 138767, 'loss/train': 1.9455629587173462}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:36 - INFO - __main__ - Step 138772: {'lr': 7.068240773208695e-06, 'samples': 26644224, 'steps': 138771, 'loss/train': 1.1098355054855347}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:38 - INFO - __main__ - Step 138776: {'lr': 7.063229808693867e-06, 'samples': 26644992, 'steps': 138775, 'loss/train': 0.18586891889572144}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:41 - INFO - __main__ - Step 138781: {'lr': 7.056968565984645e-06, 'samples': 26645952, 'steps': 138780, 'loss/train': 1.0892233848571777}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:41 - INFO - __main__ - Step 138781: {'lr': 7.056968565984645e-06, 'samples': 26645952, 'steps': 138780, 'loss/train': 1.0892233848571777}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:44 - INFO - __main__ - Step 138788: {'lr': 7.0482074237955825e-06, 'samples': 26647296, 'steps': 138787, 'loss/train': 0.668626606464386}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:46 - INFO - __main__ - Step 138792: {'lr': 7.04320346516521e-06, 'samples': 26648064, 'steps': 138791, 'loss/train': 0.8878816962242126}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:49 - INFO - __main__ - Step 138797: {'lr': 7.0369509800146395e-06, 'samples': 26649024, 'steps': 138796, 'loss/train': 1.2303940057754517}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:51 - INFO - __main__ - Step 138801: {'lr': 7.031950962447992e-06, 'samples': 26649792, 'steps': 138800, 'loss/train': 0.6443910598754883}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:51 - INFO - __main__ - Step 138801: {'lr': 7.031950962447992e-06, 'samples': 26649792, 'steps': 138800, 'loss/train': 0.6443910598754883}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:54 - INFO - __main__ - Step 138808: {'lr': 7.023205146617667e-06, 'samples': 26651136, 'steps': 138807, 'loss/train': 0.21393784880638123}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:56 - INFO - __main__ - Step 138812: {'lr': 7.0182099461545135e-06, 'samples': 26651904, 'steps': 138811, 'loss/train': 1.0326876640319824}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:47:59 - INFO - __main__ - Step 138817: {'lr': 7.011968408966258e-06, 'samples': 26652864, 'steps': 138816, 'loss/train': 1.454489827156067}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:01 - INFO - __main__ - Step 138821: {'lr': 7.006977149972088e-06, 'samples': 26653632, 'steps': 138820, 'loss/train': 1.4989750385284424}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:01 - INFO - __main__ - Step 138821: {'lr': 7.006977149972088e-06, 'samples': 26653632, 'steps': 138820, 'loss/train': 1.4989750385284424}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:04 - INFO - __main__ - Step 138828: {'lr': 6.998246662076907e-06, 'samples': 26654976, 'steps': 138827, 'loss/train': 0.6781278848648071}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:07 - INFO - __main__ - Step 138833: {'lr': 6.99201388406856e-06, 'samples': 26655936, 'steps': 138832, 'loss/train': 1.272420048713684}1}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:09 - INFO - __main__ - Step 138837: {'lr': 6.987029632580216e-06, 'samples': 26656704, 'steps': 138836, 'loss/train': 0.7104277610778809}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:09 - INFO - __main__ - Step 138837: {'lr': 6.987029632580216e-06, 'samples': 26656704, 'steps': 138836, 'loss/train': 0.7104277610778809}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:12 - INFO - __main__ - Step 138844: {'lr': 6.978311408166127e-06, 'samples': 26658048, 'steps': 138843, 'loss/train': 1.184097409248352}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:15 - INFO - __main__ - Step 138849: {'lr': 6.972087390056697e-06, 'samples': 26659008, 'steps': 138848, 'loss/train': 1.1501810550689697}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:15 - INFO - __main__ - Step 138849: {'lr': 6.972087390056697e-06, 'samples': 26659008, 'steps': 138848, 'loss/train': 1.1501810550689697}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:19 - INFO - __main__ - Step 138856: {'lr': 6.963378363913503e-06, 'samples': 26660352, 'steps': 138855, 'loss/train': 1.44522225856781}7}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:20 - INFO - __main__ - Step 138860: {'lr': 6.958404186721779e-06, 'samples': 26661120, 'steps': 138859, 'loss/train': 1.1839289665222168}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:23 - INFO - __main__ - Step 138864: {'lr': 6.953431761702711e-06, 'samples': 26661888, 'steps': 138863, 'loss/train': 1.3062798976898193}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:25 - INFO - __main__ - Step 138868: {'lr': 6.948461088892188e-06, 'samples': 26662656, 'steps': 138867, 'loss/train': 1.3625062704086304}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:27 - INFO - __main__ - Step 138872: {'lr': 6.943492168326043e-06, 'samples': 26663424, 'steps': 138871, 'loss/train': 0.6751130223274231}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:29 - INFO - __main__ - Step 138876: {'lr': 6.9385250000400526e-06, 'samples': 26664192, 'steps': 138875, 'loss/train': 0.9539720416069031}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:30 - INFO - __main__ - Step 138880: {'lr': 6.933559584070076e-06, 'samples': 26664960, 'steps': 138879, 'loss/train': 1.6544498205184937}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:33 - INFO - __main__ - Step 138884: {'lr': 6.928595920451919e-06, 'samples': 26665728, 'steps': 138883, 'loss/train': 1.3405619859695435}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:33 - INFO - __main__ - Step 138884: {'lr': 6.928595920451919e-06, 'samples': 26665728, 'steps': 138883, 'loss/train': 1.3405619859695435}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:37 - INFO - __main__ - Step 138892: {'lr': 6.918673850414087e-06, 'samples': 26667264, 'steps': 138891, 'loss/train': 0.8529251217842102}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:38 - INFO - __main__ - Step 138896: {'lr': 6.913715444065937e-06, 'samples': 26668032, 'steps': 138895, 'loss/train': 1.7336547374725342}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:41 - INFO - __main__ - Step 138900: {'lr': 6.90875879021266e-06, 'samples': 26668800, 'steps': 138899, 'loss/train': 0.9563473463058472}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:43 - INFO - __main__ - Step 138905: {'lr': 6.902565437396424e-06, 'samples': 26669760, 'steps': 138904, 'loss/train': 0.4128488302230835}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:43 - INFO - __main__ - Step 138905: {'lr': 6.902565437396424e-06, 'samples': 26669760, 'steps': 138904, 'loss/train': 0.4128488302230835}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:46 - INFO - __main__ - Step 138912: {'lr': 6.893899343979299e-06, 'samples': 26671104, 'steps': 138911, 'loss/train': 1.695116639137268}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:48 - INFO - __main__ - Step 138916: {'lr': 6.888949700462693e-06, 'samples': 26671872, 'steps': 138915, 'loss/train': 1.1775174140930176}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:51 - INFO - __main__ - Step 138921: {'lr': 6.8827651107680745e-06, 'samples': 26672832, 'steps': 138920, 'loss/train': 1.1864116191864014}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:51 - INFO - __main__ - Step 138921: {'lr': 6.8827651107680745e-06, 'samples': 26672832, 'steps': 138920, 'loss/train': 1.1864116191864014}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:54 - INFO - __main__ - Step 138928: {'lr': 6.874111286095913e-06, 'samples': 26674176, 'steps': 138927, 'loss/train': 0.060132645070552826}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:56 - INFO - __main__ - Step 138932: {'lr': 6.8691686534869126e-06, 'samples': 26674944, 'steps': 138931, 'loss/train': 0.5849564671516418}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:48:59 - INFO - __main__ - Step 138937: {'lr': 6.862992827627312e-06, 'samples': 26675904, 'steps': 138936, 'loss/train': 1.173466444015503}8}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:01 - INFO - __main__ - Step 138941: {'lr': 6.8580541389043905e-06, 'samples': 26676672, 'steps': 138940, 'loss/train': 0.8842095136642456}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:03 - INFO - __main__ - Step 138945: {'lr': 6.853117203077658e-06, 'samples': 26677440, 'steps': 138944, 'loss/train': 1.388197898864746}6}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:03 - INFO - __main__ - Step 138945: {'lr': 6.853117203077658e-06, 'samples': 26677440, 'steps': 138944, 'loss/train': 1.388197898864746}6}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:06 - INFO - __main__ - Step 138952: {'lr': 6.8444817833941684e-06, 'samples': 26678784, 'steps': 138951, 'loss/train': 1.1983423233032227}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:09 - INFO - __main__ - Step 138957: {'lr': 6.838316913330062e-06, 'samples': 26679744, 'steps': 138956, 'loss/train': 1.6521360874176025}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:11 - INFO - __main__ - Step 138962: {'lr': 6.832154782388466e-06, 'samples': 26680704, 'steps': 138961, 'loss/train': 1.043505072593689}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:11 - INFO - __main__ - Step 138962: {'lr': 6.832154782388466e-06, 'samples': 26680704, 'steps': 138961, 'loss/train': 1.043505072593689}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:15 - INFO - __main__ - Step 138969: {'lr': 6.823532400928284e-06, 'samples': 26682048, 'steps': 138968, 'loss/train': 1.4626972675323486}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:17 - INFO - __main__ - Step 138973: {'lr': 6.818607736370386e-06, 'samples': 26682816, 'steps': 138972, 'loss/train': 0.047513194382190704}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:19 - INFO - __main__ - Step 138977: {'lr': 6.813684824993005e-06, 'samples': 26683584, 'steps': 138976, 'loss/train': 0.9987470507621765}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:21 - INFO - __main__ - Step 138981: {'lr': 6.808763666831641e-06, 'samples': 26684352, 'steps': 138980, 'loss/train': 1.0106613636016846}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:21 - INFO - __main__ - Step 138981: {'lr': 6.808763666831641e-06, 'samples': 26684352, 'steps': 138980, 'loss/train': 1.0106613636016846}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:25 - INFO - __main__ - Step 138988: {'lr': 6.800155858832008e-06, 'samples': 26685696, 'steps': 138987, 'loss/train': 1.405226707458496}}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:27 - INFO - __main__ - Step 138993: {'lr': 6.7940107119983665e-06, 'samples': 26686656, 'steps': 138992, 'loss/train': 1.119428277015686}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:29 - INFO - __main__ - Step 138997: {'lr': 6.78909656705573e-06, 'samples': 26687424, 'steps': 138996, 'loss/train': 0.6626763939857483}}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:31 - INFO - __main__ - Step 139001: {'lr': 6.784184175506358e-06, 'samples': 26688192, 'steps': 139000, 'loss/train': 1.1389845609664917}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:33 - INFO - __main__ - Step 139005: {'lr': 6.7792735373856936e-06, 'samples': 26688960, 'steps': 139004, 'loss/train': 1.201503872871399}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:35 - INFO - __main__ - Step 139009: {'lr': 6.774364652729098e-06, 'samples': 26689728, 'steps': 139008, 'loss/train': 1.2672656774520874}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:35 - INFO - __main__ - Step 139009: {'lr': 6.774364652729098e-06, 'samples': 26689728, 'steps': 139008, 'loss/train': 1.2672656774520874}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:39 - INFO - __main__ - Step 139017: {'lr': 6.7645521439498316e-06, 'samples': 26691264, 'steps': 139016, 'loss/train': 0.6619526743888855}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:41 - INFO - __main__ - Step 139021: {'lr': 6.759648519897826e-06, 'samples': 26692032, 'steps': 139020, 'loss/train': 1.7338414192199707}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:42 - INFO - __main__ - Step 139025: {'lr': 6.754746649451443e-06, 'samples': 26692800, 'steps': 139024, 'loss/train': 1.496820092201233}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:45 - INFO - __main__ - Step 139029: {'lr': 6.749846532645959e-06, 'samples': 26693568, 'steps': 139028, 'loss/train': 1.3054358959197998}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:47 - INFO - __main__ - Step 139034: {'lr': 6.743723852750461e-06, 'samples': 26694528, 'steps': 139033, 'loss/train': 1.523693323135376}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:49 - INFO - __main__ - Step 139039: {'lr': 6.737603913043094e-06, 'samples': 26695488, 'steps': 139038, 'loss/train': 0.9723314642906189}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:49 - INFO - __main__ - Step 139039: {'lr': 6.737603913043094e-06, 'samples': 26695488, 'steps': 139038, 'loss/train': 0.9723314642906189}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:53 - INFO - __main__ - Step 139046: {'lr': 6.729040601100022e-06, 'samples': 26696832, 'steps': 139045, 'loss/train': 2.039492607116699}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:55 - INFO - __main__ - Step 139050: {'lr': 6.724149691488657e-06, 'samples': 26697600, 'steps': 139049, 'loss/train': 1.1761445999145508}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:57 - INFO - __main__ - Step 139054: {'lr': 6.719260535738819e-06, 'samples': 26698368, 'steps': 139053, 'loss/train': 1.3707233667373657}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:49:57 - INFO - __main__ - Step 139054: {'lr': 6.719260535738819e-06, 'samples': 26698368, 'steps': 139053, 'loss/train': 1.3707233667373657}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:01 - INFO - __main__ - Step 139062: {'lr': 6.7094874859645885e-06, 'samples': 26699904, 'steps': 139061, 'loss/train': 1.2971171140670776}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:03 - INFO - __main__ - Step 139066: {'lr': 6.704603592010694e-06, 'samples': 26700672, 'steps': 139065, 'loss/train': 0.8102603554725647}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:05 - INFO - __main__ - Step 139071: {'lr': 6.698501191138351e-06, 'samples': 26701632, 'steps': 139070, 'loss/train': 1.026437759399414}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:08 - INFO - __main__ - Step 139076: {'lr': 6.69240153096351e-06, 'samples': 26702592, 'steps': 139075, 'loss/train': 1.0483795404434204}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:08 - INFO - __main__ - Step 139076: {'lr': 6.69240153096351e-06, 'samples': 26702592, 'steps': 139075, 'loss/train': 1.0483795404434204}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:11 - INFO - __main__ - Step 139083: {'lr': 6.6838666112214176e-06, 'samples': 26703936, 'steps': 139082, 'loss/train': 0.8768734931945801}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:13 - INFO - __main__ - Step 139087: {'lr': 6.678991926172745e-06, 'samples': 26704704, 'steps': 139086, 'loss/train': 1.1754564046859741}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:16 - INFO - __main__ - Step 139092: {'lr': 6.672901036691575e-06, 'samples': 26705664, 'steps': 139091, 'loss/train': 1.2811455726623535}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:18 - INFO - __main__ - Step 139096: {'lr': 6.6680302986132094e-06, 'samples': 26706432, 'steps': 139095, 'loss/train': 0.15210847556591034}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:20 - INFO - __main__ - Step 139100: {'lr': 6.663161314800909e-06, 'samples': 26707200, 'steps': 139099, 'loss/train': 1.3453770875930786}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:22 - INFO - __main__ - Step 139104: {'lr': 6.658294085289784e-06, 'samples': 26707968, 'steps': 139103, 'loss/train': 1.0408086776733398}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:23 - INFO - __main__ - Step 139108: {'lr': 6.653428610114975e-06, 'samples': 26708736, 'steps': 139107, 'loss/train': 1.0976483821868896}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:26 - INFO - __main__ - Step 139112: {'lr': 6.648564889311509e-06, 'samples': 26709504, 'steps': 139111, 'loss/train': 1.45720636844635}6}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:28 - INFO - __main__ - Step 139117: {'lr': 6.642487705445344e-06, 'samples': 26710464, 'steps': 139116, 'loss/train': 1.3987326622009277}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:28 - INFO - __main__ - Step 139117: {'lr': 6.642487705445344e-06, 'samples': 26710464, 'steps': 139116, 'loss/train': 1.3987326622009277}4}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:31 - INFO - __main__ - Step 139124: {'lr': 6.6339842534798855e-06, 'samples': 26711808, 'steps': 139123, 'loss/train': 1.3828904628753662}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:33 - INFO - __main__ - Step 139128: {'lr': 6.629127550512398e-06, 'samples': 26712576, 'steps': 139127, 'loss/train': 1.4525502920150757}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:35 - INFO - __main__ - Step 139132: {'lr': 6.624272602091447e-06, 'samples': 26713344, 'steps': 139131, 'loss/train': 1.0217946767807007}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:35 - INFO - __main__ - Step 139132: {'lr': 6.624272602091447e-06, 'samples': 26713344, 'steps': 139131, 'loss/train': 1.0217946767807007}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:39 - INFO - __main__ - Step 139139: {'lr': 6.615780664337684e-06, 'samples': 26714688, 'steps': 139138, 'loss/train': 1.551082730293274}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:41 - INFO - __main__ - Step 139143: {'lr': 6.610930541100208e-06, 'samples': 26715456, 'steps': 139142, 'loss/train': 1.396134853363037}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:41 - INFO - __main__ - Step 139143: {'lr': 6.610930541100208e-06, 'samples': 26715456, 'steps': 139142, 'loss/train': 1.396134853363037}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:46 - INFO - __main__ - Step 139152: {'lr': 6.60002417940958e-06, 'samples': 26717184, 'steps': 139151, 'loss/train': 0.4465072751045227}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:48 - INFO - __main__ - Step 139156: {'lr': 6.595179759002534e-06, 'samples': 26717952, 'steps': 139155, 'loss/train': 1.6443133354187012}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:49 - INFO - __main__ - Step 139160: {'lr': 6.590337093386772e-06, 'samples': 26718720, 'steps': 139159, 'loss/train': 1.10319185256958}2}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:51 - INFO - __main__ - Step 139164: {'lr': 6.585496182597239e-06, 'samples': 26719488, 'steps': 139163, 'loss/train': 1.3750147819519043}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:54 - INFO - __main__ - Step 139169: {'lr': 6.579447511887826e-06, 'samples': 26720448, 'steps': 139168, 'loss/train': 1.3018754720687866}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:56 - INFO - __main__ - Step 139173: {'lr': 6.574610549584831e-06, 'samples': 26721216, 'steps': 139172, 'loss/train': 1.2482117414474487}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:56 - INFO - __main__ - Step 139173: {'lr': 6.574610549584831e-06, 'samples': 26721216, 'steps': 139172, 'loss/train': 1.2482117414474487}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:50:59 - INFO - __main__ - Step 139180: {'lr': 6.5661500883989476e-06, 'samples': 26722560, 'steps': 139179, 'loss/train': 1.0572714805603027}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:01 - INFO - __main__ - Step 139185: {'lr': 6.560110192452812e-06, 'samples': 26723520, 'steps': 139184, 'loss/train': 1.2801693677902222}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:04 - INFO - __main__ - Step 139190: {'lr': 6.554073038763097e-06, 'samples': 26724480, 'steps': 139189, 'loss/train': 1.3455939292907715}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:06 - INFO - __main__ - Step 139194: {'lr': 6.5492452902816415e-06, 'samples': 26725248, 'steps': 139193, 'loss/train': 1.2564589977264404}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:08 - INFO - __main__ - Step 139198: {'lr': 6.5444192969226765e-06, 'samples': 26726016, 'steps': 139197, 'loss/train': 1.1104450225830078}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:08 - INFO - __main__ - Step 139198: {'lr': 6.5444192969226765e-06, 'samples': 26726016, 'steps': 139197, 'loss/train': 1.1104450225830078}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:11 - INFO - __main__ - Step 139205: {'lr': 6.535978031912526e-06, 'samples': 26727360, 'steps': 139204, 'loss/train': 1.1832668781280518}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:14 - INFO - __main__ - Step 139211: {'lr': 6.528746940240859e-06, 'samples': 26728512, 'steps': 139210, 'loss/train': 1.2798329591751099}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:16 - INFO - __main__ - Step 139215: {'lr': 6.523928406540341e-06, 'samples': 26729280, 'steps': 139214, 'loss/train': 1.0103024244308472}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:18 - INFO - __main__ - Step 139219: {'lr': 6.519111628144753e-06, 'samples': 26730048, 'steps': 139218, 'loss/train': 1.3376363515853882}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:18 - INFO - __main__ - Step 139219: {'lr': 6.519111628144753e-06, 'samples': 26730048, 'steps': 139218, 'loss/train': 1.3376363515853882}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:21 - INFO - __main__ - Step 139226: {'lr': 6.510686489759527e-06, 'samples': 26731392, 'steps': 139225, 'loss/train': 0.966978907585144}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:24 - INFO - __main__ - Step 139231: {'lr': 6.5046718251350335e-06, 'samples': 26732352, 'steps': 139230, 'loss/train': 1.1858773231506348}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:24 - INFO - __main__ - Step 139231: {'lr': 6.5046718251350335e-06, 'samples': 26732352, 'steps': 139230, 'loss/train': 1.1858773231506348}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:27 - INFO - __main__ - Step 139238: {'lr': 6.496255902715908e-06, 'samples': 26733696, 'steps': 139237, 'loss/train': 0.9861428737640381}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:29 - INFO - __main__ - Step 139242: {'lr': 6.491449217997436e-06, 'samples': 26734464, 'steps': 139241, 'loss/train': 1.2747485637664795}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:29 - INFO - __main__ - Step 139242: {'lr': 6.491449217997436e-06, 'samples': 26734464, 'steps': 139241, 'loss/train': 1.2747485637664795}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:34 - INFO - __main__ - Step 139250: {'lr': 6.481841115212495e-06, 'samples': 26736000, 'steps': 139249, 'loss/train': 1.1962602138519287}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:36 - INFO - __main__ - Step 139254: {'lr': 6.477039697215331e-06, 'samples': 26736768, 'steps': 139253, 'loss/train': 1.4039466381072998}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:37 - INFO - __main__ - Step 139258: {'lr': 6.4722400348611325e-06, 'samples': 26737536, 'steps': 139257, 'loss/train': 0.8238503336906433}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:40 - INFO - __main__ - Step 139263: {'lr': 6.466242925844079e-06, 'samples': 26738496, 'steps': 139262, 'loss/train': 1.0916188955307007}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:42 - INFO - __main__ - Step 139267: {'lr': 6.461447213813171e-06, 'samples': 26739264, 'steps': 139266, 'loss/train': 1.5431736707687378}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:44 - INFO - __main__ - Step 139271: {'lr': 6.456653257537665e-06, 'samples': 26740032, 'steps': 139270, 'loss/train': 1.273200273513794}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:46 - INFO - __main__ - Step 139275: {'lr': 6.451861057052144e-06, 'samples': 26740800, 'steps': 139274, 'loss/train': 1.2950797080993652}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:47 - INFO - __main__ - Step 139279: {'lr': 6.447070612391193e-06, 'samples': 26741568, 'steps': 139278, 'loss/train': 1.4411970376968384}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:49 - INFO - __main__ - Step 139283: {'lr': 6.442281923589255e-06, 'samples': 26742336, 'steps': 139282, 'loss/train': 1.6882282495498657}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:49 - INFO - __main__ - Step 139283: {'lr': 6.442281923589255e-06, 'samples': 26742336, 'steps': 139282, 'loss/train': 1.6882282495498657}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:49 - INFO - __main__ - Step 139283: {'lr': 6.442281923589255e-06, 'samples': 26742336, 'steps': 139282, 'loss/train': 1.6882282495498657}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:55 - INFO - __main__ - Step 139294: {'lr': 6.429122083314115e-06, 'samples': 26744448, 'steps': 139293, 'loss/train': 0.7931724786758423}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:51:58 - INFO - __main__ - Step 139299: {'lr': 6.423144727662534e-06, 'samples': 26745408, 'steps': 139298, 'loss/train': 1.5615119934082031}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:00 - INFO - __main__ - Step 139304: {'lr': 6.41717011581025e-06, 'samples': 26746368, 'steps': 139303, 'loss/train': 1.3191357851028442}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:02 - INFO - __main__ - Step 139308: {'lr': 6.4123924019091896e-06, 'samples': 26747136, 'steps': 139307, 'loss/train': 1.5454264879226685}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:02 - INFO - __main__ - Step 139308: {'lr': 6.4123924019091896e-06, 'samples': 26747136, 'steps': 139307, 'loss/train': 1.5454264879226685}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:05 - INFO - __main__ - Step 139315: {'lr': 6.4040356282404626e-06, 'samples': 26748480, 'steps': 139314, 'loss/train': 1.3937162160873413}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:07 - INFO - __main__ - Step 139319: {'lr': 6.3992627437221565e-06, 'samples': 26749248, 'steps': 139318, 'loss/train': 1.2908084392547607}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:10 - INFO - __main__ - Step 139324: {'lr': 6.393299107739836e-06, 'samples': 26750208, 'steps': 139323, 'loss/train': 0.9640383720397949}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:12 - INFO - __main__ - Step 139329: {'lr': 6.387338215893018e-06, 'samples': 26751168, 'steps': 139328, 'loss/train': 1.3290627002716064}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:12 - INFO - __main__ - Step 139329: {'lr': 6.387338215893018e-06, 'samples': 26751168, 'steps': 139328, 'loss/train': 1.3290627002716064}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:15 - INFO - __main__ - Step 139334: {'lr': 6.381380068248844e-06, 'samples': 26752128, 'steps': 139333, 'loss/train': 1.7348772287368774}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:15 - INFO - __main__ - Step 139334: {'lr': 6.381380068248844e-06, 'samples': 26752128, 'steps': 139333, 'loss/train': 1.7348772287368774}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:19 - INFO - __main__ - Step 139343: {'lr': 6.370662318094245e-06, 'samples': 26753856, 'steps': 139342, 'loss/train': 1.1430033445358276}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:22 - INFO - __main__ - Step 139347: {'lr': 6.365901727723972e-06, 'samples': 26754624, 'steps': 139346, 'loss/train': 1.5417646169662476}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:24 - INFO - __main__ - Step 139352: {'lr': 6.359953459764839e-06, 'samples': 26755584, 'steps': 139351, 'loss/train': 1.0465153455734253}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:24 - INFO - __main__ - Step 139352: {'lr': 6.359953459764839e-06, 'samples': 26755584, 'steps': 139351, 'loss/train': 1.0465153455734253}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:28 - INFO - __main__ - Step 139360: {'lr': 6.350441939641266e-06, 'samples': 26757120, 'steps': 139359, 'loss/train': 1.2434769868850708}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:30 - INFO - __main__ - Step 139364: {'lr': 6.3456888143959e-06, 'samples': 26757888, 'steps': 139363, 'loss/train': 1.8136459589004517}8}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:32 - INFO - __main__ - Step 139368: {'lr': 6.340937445740463e-06, 'samples': 26758656, 'steps': 139367, 'loss/train': 1.037156105041504}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:34 - INFO - __main__ - Step 139372: {'lr': 6.336187833709261e-06, 'samples': 26759424, 'steps': 139371, 'loss/train': 1.3633711338043213}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:36 - INFO - __main__ - Step 139377: {'lr': 6.330253288975241e-06, 'samples': 26760384, 'steps': 139376, 'loss/train': 1.03940749168396}3}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:38 - INFO - __main__ - Step 139381: {'lr': 6.3255076294737035e-06, 'samples': 26761152, 'steps': 139380, 'loss/train': 0.9532141089439392}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:40 - INFO - __main__ - Step 139385: {'lr': 6.320763726707618e-06, 'samples': 26761920, 'steps': 139384, 'loss/train': 1.5496069192886353}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:42 - INFO - __main__ - Step 139389: {'lr': 6.316021580711234e-06, 'samples': 26762688, 'steps': 139388, 'loss/train': 1.322221040725708}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:44 - INFO - __main__ - Step 139393: {'lr': 6.311281191518637e-06, 'samples': 26763456, 'steps': 139392, 'loss/train': 1.0670835971832275}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:44 - INFO - __main__ - Step 139393: {'lr': 6.311281191518637e-06, 'samples': 26763456, 'steps': 139392, 'loss/train': 1.0670835971832275}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:49 - INFO - __main__ - Step 139401: {'lr': 6.301805683681744e-06, 'samples': 26764992, 'steps': 139400, 'loss/train': 1.3751665353775024}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:49 - INFO - __main__ - Step 139401: {'lr': 6.301805683681744e-06, 'samples': 26764992, 'steps': 139400, 'loss/train': 1.3751665353775024}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:52 - INFO - __main__ - Step 139408: {'lr': 6.293520379164003e-06, 'samples': 26766336, 'steps': 139407, 'loss/train': 1.2145891189575195}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:54 - INFO - __main__ - Step 139412: {'lr': 6.288788335256174e-06, 'samples': 26767104, 'steps': 139411, 'loss/train': 1.307380199432373}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:56 - INFO - __main__ - Step 139416: {'lr': 6.284058048348529e-06, 'samples': 26767872, 'steps': 139415, 'loss/train': 1.6341912746429443}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:52:58 - INFO - __main__ - Step 139420: {'lr': 6.279329518475207e-06, 'samples': 26768640, 'steps': 139419, 'loss/train': 1.0262486934661865}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:00 - INFO - __main__ - Step 139424: {'lr': 6.274602745670211e-06, 'samples': 26769408, 'steps': 139423, 'loss/train': 1.0869041681289673}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:02 - INFO - __main__ - Step 139429: {'lr': 6.268696750593272e-06, 'samples': 26770368, 'steps': 139428, 'loss/train': 1.3181859254837036}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:02 - INFO - __main__ - Step 139429: {'lr': 6.268696750593272e-06, 'samples': 26770368, 'steps': 139428, 'loss/train': 1.3181859254837036}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:06 - INFO - __main__ - Step 139436: {'lr': 6.260432970006197e-06, 'samples': 26771712, 'steps': 139435, 'loss/train': 1.1248836517333984}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:08 - INFO - __main__ - Step 139440: {'lr': 6.2557132258153345e-06, 'samples': 26772480, 'steps': 139439, 'loss/train': 0.9527719020843506}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:10 - INFO - __main__ - Step 139444: {'lr': 6.250995238863133e-06, 'samples': 26773248, 'steps': 139443, 'loss/train': 1.5830364227294922}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:12 - INFO - __main__ - Step 139449: {'lr': 6.245100226341477e-06, 'samples': 26774208, 'steps': 139448, 'loss/train': 1.0579943656921387}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:14 - INFO - __main__ - Step 139453: {'lr': 6.240386193300502e-06, 'samples': 26774976, 'steps': 139452, 'loss/train': 1.3164448738098145}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:16 - INFO - __main__ - Step 139457: {'lr': 6.235673917608681e-06, 'samples': 26775744, 'steps': 139456, 'loss/train': 1.2546581029891968}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:18 - INFO - __main__ - Step 139461: {'lr': 6.230963399299933e-06, 'samples': 26776512, 'steps': 139460, 'loss/train': 1.2280235290527344}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:20 - INFO - __main__ - Step 139465: {'lr': 6.226254638408257e-06, 'samples': 26777280, 'steps': 139464, 'loss/train': 1.2733012437820435}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:22 - INFO - __main__ - Step 139470: {'lr': 6.2203711587131284e-06, 'samples': 26778240, 'steps': 139469, 'loss/train': 1.2935926914215088}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:24 - INFO - __main__ - Step 139474: {'lr': 6.215666352133914e-06, 'samples': 26779008, 'steps': 139473, 'loss/train': 1.3835285902023315}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:24 - INFO - __main__ - Step 139474: {'lr': 6.215666352133914e-06, 'samples': 26779008, 'steps': 139473, 'loss/train': 1.3835285902023315}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:27 - INFO - __main__ - Step 139481: {'lr': 6.20743716969091e-06, 'samples': 26780352, 'steps': 139480, 'loss/train': 1.1649107933044434}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:30 - INFO - __main__ - Step 139485: {'lr': 6.202737196393471e-06, 'samples': 26781120, 'steps': 139484, 'loss/train': 1.584341049194336}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:32 - INFO - __main__ - Step 139490: {'lr': 6.1968647014295495e-06, 'samples': 26782080, 'steps': 139489, 'loss/train': 1.2720274925231934}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:34 - INFO - __main__ - Step 139494: {'lr': 6.192168682826016e-06, 'samples': 26782848, 'steps': 139493, 'loss/train': 0.9385887384414673}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:36 - INFO - __main__ - Step 139498: {'lr': 6.18747442191922e-06, 'samples': 26783616, 'steps': 139497, 'loss/train': 0.97804856300354}73}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:38 - INFO - __main__ - Step 139502: {'lr': 6.182781918742997e-06, 'samples': 26784384, 'steps': 139501, 'loss/train': 0.8258057832717896}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:40 - INFO - __main__ - Step 139507: {'lr': 6.1769187616328715e-06, 'samples': 26785344, 'steps': 139506, 'loss/train': 1.4368401765823364}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:43 - INFO - __main__ - Step 139511: {'lr': 6.172230213474156e-06, 'samples': 26786112, 'steps': 139510, 'loss/train': 1.3521976470947266}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:43 - INFO - __main__ - Step 139511: {'lr': 6.172230213474156e-06, 'samples': 26786112, 'steps': 139510, 'loss/train': 1.3521976470947266}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:46 - INFO - __main__ - Step 139518: {'lr': 6.1640294840203944e-06, 'samples': 26787456, 'steps': 139517, 'loss/train': 1.436779260635376}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:48 - INFO - __main__ - Step 139522: {'lr': 6.159345770004321e-06, 'samples': 26788224, 'steps': 139521, 'loss/train': 1.3060152530670166}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:50 - INFO - __main__ - Step 139527: {'lr': 6.1534935995819775e-06, 'samples': 26789184, 'steps': 139526, 'loss/train': 1.5999687910079956}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:52 - INFO - __main__ - Step 139531: {'lr': 6.148813840963441e-06, 'samples': 26789952, 'steps': 139530, 'loss/train': 0.29953062534332275}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:55 - INFO - __main__ - Step 139535: {'lr': 6.1441358403542255e-06, 'samples': 26790720, 'steps': 139534, 'loss/train': 1.2131303548812866}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:56 - INFO - __main__ - Step 139539: {'lr': 6.139459597788027e-06, 'samples': 26791488, 'steps': 139538, 'loss/train': 1.6483235359191895}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:58 - INFO - __main__ - Step 139543: {'lr': 6.134785113298569e-06, 'samples': 26792256, 'steps': 139542, 'loss/train': 1.3131773471832275}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:53:58 - INFO - __main__ - Step 139543: {'lr': 6.134785113298569e-06, 'samples': 26792256, 'steps': 139542, 'loss/train': 1.3131773471832275}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:03 - INFO - __main__ - Step 139551: {'lr': 6.1254414186847075e-06, 'samples': 26793792, 'steps': 139550, 'loss/train': 1.4869277477264404}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:03 - INFO - __main__ - Step 139551: {'lr': 6.1254414186847075e-06, 'samples': 26793792, 'steps': 139550, 'loss/train': 1.4869277477264404}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:06 - INFO - __main__ - Step 139558: {'lr': 6.117271454909257e-06, 'samples': 26795136, 'steps': 139557, 'loss/train': 0.231853649020195}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:09 - INFO - __main__ - Step 139564: {'lr': 6.110272914511189e-06, 'samples': 26796288, 'steps': 139563, 'loss/train': 1.208053469657898}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:09 - INFO - __main__ - Step 139564: {'lr': 6.110272914511189e-06, 'samples': 26796288, 'steps': 139563, 'loss/train': 1.208053469657898}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:13 - INFO - __main__ - Step 139571: {'lr': 6.102112950850475e-06, 'samples': 26797632, 'steps': 139570, 'loss/train': 0.6476883888244629}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:14 - INFO - __main__ - Step 139575: {'lr': 6.0974525321871035e-06, 'samples': 26798400, 'steps': 139574, 'loss/train': 1.2892582416534424}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:16 - INFO - __main__ - Step 139579: {'lr': 6.09279387190334e-06, 'samples': 26799168, 'steps': 139578, 'loss/train': 1.3000683784484863}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:16 - INFO - __main__ - Step 139579: {'lr': 6.09279387190334e-06, 'samples': 26799168, 'steps': 139578, 'loss/train': 1.3000683784484863}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:21 - INFO - __main__ - Step 139587: {'lr': 6.083481826608839e-06, 'samples': 26800704, 'steps': 139586, 'loss/train': 0.036827582865953445}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:22 - INFO - __main__ - Step 139591: {'lr': 6.0788284416653235e-06, 'samples': 26801472, 'steps': 139590, 'loss/train': 0.8717607855796814}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:24 - INFO - __main__ - Step 139595: {'lr': 6.074176815235615e-06, 'samples': 26802240, 'steps': 139594, 'loss/train': 1.4250543117523193}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:26 - INFO - __main__ - Step 139600: {'lr': 6.068364755159678e-06, 'samples': 26803200, 'steps': 139599, 'loss/train': 1.114575743675232}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:29 - INFO - __main__ - Step 139604: {'lr': 6.063717085508763e-06, 'samples': 26803968, 'steps': 139603, 'loss/train': 0.4777734875679016}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:31 - INFO - __main__ - Step 139608: {'lr': 6.059071174480623e-06, 'samples': 26804736, 'steps': 139607, 'loss/train': 1.138903260231018}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:31 - INFO - __main__ - Step 139608: {'lr': 6.059071174480623e-06, 'samples': 26804736, 'steps': 139607, 'loss/train': 1.138903260231018}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:34 - INFO - __main__ - Step 139615: {'lr': 6.050945061968238e-06, 'samples': 26806080, 'steps': 139614, 'loss/train': 0.8265625238418579}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:36 - INFO - __main__ - Step 139620: {'lr': 6.045143993467867e-06, 'samples': 26807040, 'steps': 139619, 'loss/train': 1.380424976348877}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:39 - INFO - __main__ - Step 139625: {'lr': 6.0393456730248595e-06, 'samples': 26808000, 'steps': 139624, 'loss/train': 1.1686195135116577}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:41 - INFO - __main__ - Step 139630: {'lr': 6.033550100704526e-06, 'samples': 26808960, 'steps': 139629, 'loss/train': 1.0188207626342773}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:41 - INFO - __main__ - Step 139630: {'lr': 6.033550100704526e-06, 'samples': 26808960, 'steps': 139629, 'loss/train': 1.0188207626342773}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:44 - INFO - __main__ - Step 139636: {'lr': 6.026599041533909e-06, 'samples': 26810112, 'steps': 139635, 'loss/train': 1.176435112953186}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:44 - INFO - __main__ - Step 139636: {'lr': 6.026599041533909e-06, 'samples': 26810112, 'steps': 139635, 'loss/train': 1.176435112953186}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:49 - INFO - __main__ - Step 139644: {'lr': 6.01733711877564e-06, 'samples': 26811648, 'steps': 139643, 'loss/train': 0.8666209578514099}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:50 - INFO - __main__ - Step 139648: {'lr': 6.01270879581553e-06, 'samples': 26812416, 'steps': 139647, 'loss/train': 1.2959458827972412}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:52 - INFO - __main__ - Step 139652: {'lr': 6.008082231845985e-06, 'samples': 26813184, 'steps': 139651, 'loss/train': 1.2207046747207642}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:55 - INFO - __main__ - Step 139657: {'lr': 6.002301500515339e-06, 'samples': 26814144, 'steps': 139656, 'loss/train': 1.229170560836792}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:55 - INFO - __main__ - Step 139657: {'lr': 6.002301500515339e-06, 'samples': 26814144, 'steps': 139656, 'loss/train': 1.229170560836792}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:54:58 - INFO - __main__ - Step 139664: {'lr': 5.9942130942140516e-06, 'samples': 26815488, 'steps': 139663, 'loss/train': 0.9343996047973633}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:00 - INFO - __main__ - Step 139668: {'lr': 5.98959356654008e-06, 'samples': 26816256, 'steps': 139667, 'loss/train': 1.2403289079666138}3}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:04 - INFO - __main__ - Step 139673: {'lr': 5.983821630766273e-06, 'samples': 26817216, 'steps': 139672, 'loss/train': 0.9560865759849548}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:07 - INFO - __main__ - Step 139679: {'lr': 5.976898936190939e-06, 'samples': 26818368, 'steps': 139678, 'loss/train': 1.5525866746902466}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:08 - INFO - __main__ - Step 139683: {'lr': 5.972286005527988e-06, 'samples': 26819136, 'steps': 139682, 'loss/train': 1.4552273750305176}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:10 - INFO - __main__ - Step 139687: {'lr': 5.967674834147035e-06, 'samples': 26819904, 'steps': 139686, 'loss/train': 1.5676593780517578}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:13 - INFO - __main__ - Step 139692: {'lr': 5.961913343961678e-06, 'samples': 26820864, 'steps': 139691, 'loss/train': 1.486158847808838}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:13 - INFO - __main__ - Step 139692: {'lr': 5.961913343961678e-06, 'samples': 26820864, 'steps': 139691, 'loss/train': 1.486158847808838}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:16 - INFO - __main__ - Step 139699: {'lr': 5.953851876028177e-06, 'samples': 26822208, 'steps': 139698, 'loss/train': 1.0661251544952393}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:19 - INFO - __main__ - Step 139703: {'lr': 5.949247742107311e-06, 'samples': 26822976, 'steps': 139702, 'loss/train': 1.7984400987625122}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:21 - INFO - __main__ - Step 139707: {'lr': 5.9446453676345045e-06, 'samples': 26823744, 'steps': 139706, 'loss/train': 1.218497395515442}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:23 - INFO - __main__ - Step 139711: {'lr': 5.9400447526429535e-06, 'samples': 26824512, 'steps': 139710, 'loss/train': 1.3231757879257202}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:24 - INFO - __main__ - Step 139715: {'lr': 5.93544589716577e-06, 'samples': 26825280, 'steps': 139714, 'loss/train': 1.2119572162628174}2}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:26 - INFO - __main__ - Step 139719: {'lr': 5.930848801236122e-06, 'samples': 26826048, 'steps': 139718, 'loss/train': 1.0811388492584229}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:29 - INFO - __main__ - Step 139724: {'lr': 5.925104905738332e-06, 'samples': 26827008, 'steps': 139723, 'loss/train': 1.0583826303482056}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:31 - INFO - __main__ - Step 139728: {'lr': 5.92051176891184e-06, 'samples': 26827776, 'steps': 139727, 'loss/train': 1.468629240989685}6}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:33 - INFO - __main__ - Step 139732: {'lr': 5.915920391740548e-06, 'samples': 26828544, 'steps': 139731, 'loss/train': 1.5811108350753784}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:33 - INFO - __main__ - Step 139732: {'lr': 5.915920391740548e-06, 'samples': 26828544, 'steps': 139731, 'loss/train': 1.5811108350753784}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:36 - INFO - __main__ - Step 139739: {'lr': 5.907889715960762e-06, 'samples': 26829888, 'steps': 139738, 'loss/train': 1.2298741340637207}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:39 - INFO - __main__ - Step 139744: {'lr': 5.90215681848899e-06, 'samples': 26830848, 'steps': 139743, 'loss/train': 1.3857921361923218}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:41 - INFO - __main__ - Step 139749: {'lr': 5.89642667068524e-06, 'samples': 26831808, 'steps': 139748, 'loss/train': 1.3923285007476807}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:41 - INFO - __main__ - Step 139749: {'lr': 5.89642667068524e-06, 'samples': 26831808, 'steps': 139748, 'loss/train': 1.3923285007476807}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:44 - INFO - __main__ - Step 139756: {'lr': 5.888409083325219e-06, 'samples': 26833152, 'steps': 139755, 'loss/train': 1.312744140625}07}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:47 - INFO - __main__ - Step 139760: {'lr': 5.883830024666553e-06, 'samples': 26833920, 'steps': 139759, 'loss/train': 1.2822849750518799}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:49 - INFO - __main__ - Step 139764: {'lr': 5.879252725927598e-06, 'samples': 26834688, 'steps': 139763, 'loss/train': 1.5333613157272339}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:51 - INFO - __main__ - Step 139768: {'lr': 5.87467718714127e-06, 'samples': 26835456, 'steps': 139767, 'loss/train': 1.3468923568725586}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:53 - INFO - __main__ - Step 139772: {'lr': 5.870103408340599e-06, 'samples': 26836224, 'steps': 139771, 'loss/train': 1.500889539718628}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:55 - INFO - __main__ - Step 139776: {'lr': 5.865531389558559e-06, 'samples': 26836992, 'steps': 139775, 'loss/train': 1.0574229955673218}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:57 - INFO - __main__ - Step 139781: {'lr': 5.85981884115741e-06, 'samples': 26837952, 'steps': 139780, 'loss/train': 1.0949020385742188}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:55:57 - INFO - __main__ - Step 139781: {'lr': 5.85981884115741e-06, 'samples': 26837952, 'steps': 139780, 'loss/train': 1.0949020385742188}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:01 - INFO - __main__ - Step 139788: {'lr': 5.8518258936537394e-06, 'samples': 26839296, 'steps': 139787, 'loss/train': 1.0971578359603882}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:03 - INFO - __main__ - Step 139793: {'lr': 5.846119945708578e-06, 'samples': 26840256, 'steps': 139792, 'loss/train': 1.0355472564697266}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:05 - INFO - __main__ - Step 139797: {'lr': 5.8415571675647925e-06, 'samples': 26841024, 'steps': 139796, 'loss/train': 1.3677871227264404}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:05 - INFO - __main__ - Step 139797: {'lr': 5.8415571675647925e-06, 'samples': 26841024, 'steps': 139796, 'loss/train': 1.3677871227264404}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:09 - INFO - __main__ - Step 139804: {'lr': 5.8335765413730094e-06, 'samples': 26842368, 'steps': 139803, 'loss/train': 1.2783828973770142}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:11 - INFO - __main__ - Step 139808: {'lr': 5.82901860392554e-06, 'samples': 26843136, 'steps': 139807, 'loss/train': 1.3288275003433228}2}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:11 - INFO - __main__ - Step 139808: {'lr': 5.82901860392554e-06, 'samples': 26843136, 'steps': 139807, 'loss/train': 1.3288275003433228}2}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:15 - INFO - __main__ - Step 139816: {'lr': 5.819908010007868e-06, 'samples': 26844672, 'steps': 139815, 'loss/train': 1.6285799741744995}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:17 - INFO - __main__ - Step 139820: {'lr': 5.815355353603391e-06, 'samples': 26845440, 'steps': 139819, 'loss/train': 1.0797349214553833}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:19 - INFO - __main__ - Step 139824: {'lr': 5.8108044576122285e-06, 'samples': 26846208, 'steps': 139823, 'loss/train': 1.2923723459243774}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:21 - INFO - __main__ - Step 139829: {'lr': 5.805118313254476e-06, 'samples': 26847168, 'steps': 139828, 'loss/train': 1.2579503059387207}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:21 - INFO - __main__ - Step 139829: {'lr': 5.805118313254476e-06, 'samples': 26847168, 'steps': 139828, 'loss/train': 1.2579503059387207}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:24 - INFO - __main__ - Step 139836: {'lr': 5.797162332446731e-06, 'samples': 26848512, 'steps': 139835, 'loss/train': 1.4072179794311523}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:26 - INFO - __main__ - Step 139840: {'lr': 5.792618478436817e-06, 'samples': 26849280, 'steps': 139839, 'loss/train': 1.366459608078003}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:29 - INFO - __main__ - Step 139845: {'lr': 5.786941136740054e-06, 'samples': 26850240, 'steps': 139844, 'loss/train': 1.170481562614441}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:31 - INFO - __main__ - Step 139849: {'lr': 5.782401244074975e-06, 'samples': 26851008, 'steps': 139848, 'loss/train': 1.4384750127792358}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:31 - INFO - __main__ - Step 139849: {'lr': 5.782401244074975e-06, 'samples': 26851008, 'steps': 139848, 'loss/train': 1.4384750127792358}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:35 - INFO - __main__ - Step 139856: {'lr': 5.7744606684970444e-06, 'samples': 26852352, 'steps': 139855, 'loss/train': 0.9315148591995239}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:37 - INFO - __main__ - Step 139860: {'lr': 5.769925617700705e-06, 'samples': 26853120, 'steps': 139859, 'loss/train': 1.1505433320999146}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:39 - INFO - __main__ - Step 139865: {'lr': 5.764259280250899e-06, 'samples': 26854080, 'steps': 139864, 'loss/train': 1.0256165266036987}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:41 - INFO - __main__ - Step 139869: {'lr': 5.759728191167407e-06, 'samples': 26854848, 'steps': 139868, 'loss/train': 1.3498883247375488}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:41 - INFO - __main__ - Step 139869: {'lr': 5.759728191167407e-06, 'samples': 26854848, 'steps': 139868, 'loss/train': 1.3498883247375488}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:45 - INFO - __main__ - Step 139876: {'lr': 5.751803022250479e-06, 'samples': 26856192, 'steps': 139875, 'loss/train': 0.9643576741218567}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:47 - INFO - __main__ - Step 139881: {'lr': 5.746145488933679e-06, 'samples': 26857152, 'steps': 139880, 'loss/train': 1.3975803852081299}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:47 - INFO - __main__ - Step 139881: {'lr': 5.746145488933679e-06, 'samples': 26857152, 'steps': 139880, 'loss/train': 1.3975803852081299}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:51 - INFO - __main__ - Step 139889: {'lr': 5.737099158618225e-06, 'samples': 26858688, 'steps': 139888, 'loss/train': 1.1917942762374878}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:51 - INFO - __main__ - Step 139889: {'lr': 5.737099158618225e-06, 'samples': 26858688, 'steps': 139888, 'loss/train': 1.1917942762374878}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:55 - INFO - __main__ - Step 139896: {'lr': 5.7291893977906855e-06, 'samples': 26860032, 'steps': 139895, 'loss/train': 1.4626458883285522}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:57 - INFO - __main__ - Step 139901: {'lr': 5.723542870557735e-06, 'samples': 26860992, 'steps': 139900, 'loss/train': 1.0729916095733643}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:56:57 - INFO - __main__ - Step 139901: {'lr': 5.723542870557735e-06, 'samples': 26860992, 'steps': 139900, 'loss/train': 1.0729916095733643}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:01 - INFO - __main__ - Step 139908: {'lr': 5.715642355268308e-06, 'samples': 26862336, 'steps': 139907, 'loss/train': 1.5296695232391357}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:03 - INFO - __main__ - Step 139912: {'lr': 5.711130196648512e-06, 'samples': 26863104, 'steps': 139911, 'loss/train': 1.5159430503845215}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:05 - INFO - __main__ - Step 139917: {'lr': 5.7054924750154505e-06, 'samples': 26864064, 'steps': 139916, 'loss/train': 1.0802339315414429}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:08 - INFO - __main__ - Step 139922: {'lr': 5.699857505265749e-06, 'samples': 26865024, 'steps': 139921, 'loss/train': 1.2056344747543335}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:10 - INFO - __main__ - Step 139926: {'lr': 5.695351510864727e-06, 'samples': 26865792, 'steps': 139925, 'loss/train': 1.3610447645187378}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:12 - INFO - __main__ - Step 139930: {'lr': 5.690847277742134e-06, 'samples': 26866560, 'steps': 139929, 'loss/train': 1.6749238967895508}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:13 - INFO - __main__ - Step 139934: {'lr': 5.6863448059305266e-06, 'samples': 26867328, 'steps': 139933, 'loss/train': 0.8607516288757324}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:15 - INFO - __main__ - Step 139938: {'lr': 5.681844095462296e-06, 'samples': 26868096, 'steps': 139937, 'loss/train': 1.214661955833435}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:18 - INFO - __main__ - Step 139943: {'lr': 5.676220684315614e-06, 'samples': 26869056, 'steps': 139942, 'loss/train': 1.1281887292861938}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:18 - INFO - __main__ - Step 139943: {'lr': 5.676220684315614e-06, 'samples': 26869056, 'steps': 139942, 'loss/train': 1.1281887292861938}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:22 - INFO - __main__ - Step 139950: {'lr': 5.668352532442494e-06, 'samples': 26870400, 'steps': 139949, 'loss/train': 1.5413345098495483}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:23 - INFO - __main__ - Step 139954: {'lr': 5.663858867672261e-06, 'samples': 26871168, 'steps': 139953, 'loss/train': 1.109904170036316}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:25 - INFO - __main__ - Step 139958: {'lr': 5.659366964407553e-06, 'samples': 26871936, 'steps': 139957, 'loss/train': 1.2128835916519165}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:25 - INFO - __main__ - Step 139958: {'lr': 5.659366964407553e-06, 'samples': 26871936, 'steps': 139957, 'loss/train': 1.2128835916519165}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:29 - INFO - __main__ - Step 139966: {'lr': 5.650388442524162e-06, 'samples': 26873472, 'steps': 139965, 'loss/train': 1.5014739036560059}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:32 - INFO - __main__ - Step 139970: {'lr': 5.645901823970234e-06, 'samples': 26874240, 'steps': 139969, 'loss/train': 1.4964226484298706}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:33 - INFO - __main__ - Step 139974: {'lr': 5.641416967051283e-06, 'samples': 26875008, 'steps': 139973, 'loss/train': 1.2170170545578003}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:35 - INFO - __main__ - Step 139978: {'lr': 5.636933871799671e-06, 'samples': 26875776, 'steps': 139977, 'loss/train': 1.630928635597229}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:37 - INFO - __main__ - Step 139983: {'lr': 5.631332480128981e-06, 'samples': 26876736, 'steps': 139982, 'loss/train': 1.472499966621399}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:37 - INFO - __main__ - Step 139983: {'lr': 5.631332480128981e-06, 'samples': 26876736, 'steps': 139982, 'loss/train': 1.472499966621399}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:41 - INFO - __main__ - Step 139989: {'lr': 5.624614443718506e-06, 'samples': 26877888, 'steps': 139988, 'loss/train': 3.758086681365967}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:44 - INFO - __main__ - Step 139995: {'lr': 5.61790037133203e-06, 'samples': 26879040, 'steps': 139994, 'loss/train': 1.3481099605560303}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:46 - INFO - __main__ - Step 139999: {'lr': 5.613426525363729e-06, 'samples': 26879808, 'steps': 139998, 'loss/train': 1.57685124874115}}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:46 - INFO - __main__ - Step 139999: {'lr': 5.613426525363729e-06, 'samples': 26879808, 'steps': 139998, 'loss/train': 1.57685124874115}}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:49 - INFO - __main__ - Step 140006: {'lr': 5.605601534436094e-06, 'samples': 26881152, 'steps': 140005, 'loss/train': 0.7840285301208496}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:52 - INFO - __main__ - Step 140011: {'lr': 5.600015558802352e-06, 'samples': 26882112, 'steps': 140010, 'loss/train': 1.0059764385223389}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:54 - INFO - __main__ - Step 140016: {'lr': 5.594432336240129e-06, 'samples': 26883072, 'steps': 140015, 'loss/train': 1.5925344228744507}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:56 - INFO - __main__ - Step 140020: {'lr': 5.589967740444124e-06, 'samples': 26883840, 'steps': 140019, 'loss/train': 1.632413625717163}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:58 - INFO - __main__ - Step 140024: {'lr': 5.585504906686356e-06, 'samples': 26884608, 'steps': 140023, 'loss/train': 1.257991075515747}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:57:58 - INFO - __main__ - Step 140024: {'lr': 5.585504906686356e-06, 'samples': 26884608, 'steps': 140023, 'loss/train': 1.257991075515747}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:01 - INFO - __main__ - Step 140031: {'lr': 5.577699187611474e-06, 'samples': 26885952, 'steps': 140030, 'loss/train': 1.611396074295044}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:04 - INFO - __main__ - Step 140036: {'lr': 5.572126977964053e-06, 'samples': 26886912, 'steps': 140035, 'loss/train': 0.969295859336853}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:06 - INFO - __main__ - Step 140040: {'lr': 5.567671192680785e-06, 'samples': 26887680, 'steps': 140039, 'loss/train': 1.467738151550293}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:08 - INFO - __main__ - Step 140044: {'lr': 5.563217169596485e-06, 'samples': 26888448, 'steps': 140043, 'loss/train': 1.506882667541504}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:09 - INFO - __main__ - Step 140048: {'lr': 5.558764908743269e-06, 'samples': 26889216, 'steps': 140047, 'loss/train': 1.3863661289215088}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:11 - INFO - __main__ - Step 140052: {'lr': 5.554314410153221e-06, 'samples': 26889984, 'steps': 140051, 'loss/train': 0.9723540544509888}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:14 - INFO - __main__ - Step 140057: {'lr': 5.54875376514713e-06, 'samples': 26890944, 'steps': 140056, 'loss/train': 1.2341680526733398}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:16 - INFO - __main__ - Step 140061: {'lr': 5.544307231766549e-06, 'samples': 26891712, 'steps': 140060, 'loss/train': 0.9828872680664062}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:16 - INFO - __main__ - Step 140061: {'lr': 5.544307231766549e-06, 'samples': 26891712, 'steps': 140060, 'loss/train': 0.9828872680664062}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:19 - INFO - __main__ - Step 140068: {'lr': 5.536530039066317e-06, 'samples': 26893056, 'steps': 140067, 'loss/train': 1.10368013381958}2}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:22 - INFO - __main__ - Step 140073: {'lr': 5.5309782059573544e-06, 'samples': 26894016, 'steps': 140072, 'loss/train': 1.299936056137085}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:24 - INFO - __main__ - Step 140077: {'lr': 5.526538722238572e-06, 'samples': 26894784, 'steps': 140076, 'loss/train': 1.2396804094314575}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:24 - INFO - __main__ - Step 140077: {'lr': 5.526538722238572e-06, 'samples': 26894784, 'steps': 140076, 'loss/train': 1.2396804094314575}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:27 - INFO - __main__ - Step 140084: {'lr': 5.518773866754784e-06, 'samples': 26896128, 'steps': 140083, 'loss/train': 1.9322186708450317}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:30 - INFO - __main__ - Step 140089: {'lr': 5.51323084618327e-06, 'samples': 26897088, 'steps': 140088, 'loss/train': 0.8608591556549072}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:32 - INFO - __main__ - Step 140093: {'lr': 5.508798412638433e-06, 'samples': 26897856, 'steps': 140092, 'loss/train': 0.9306735396385193}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:32 - INFO - __main__ - Step 140093: {'lr': 5.508798412638433e-06, 'samples': 26897856, 'steps': 140092, 'loss/train': 0.9306735396385193}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:36 - INFO - __main__ - Step 140101: {'lr': 5.499938833450929e-06, 'samples': 26899392, 'steps': 140100, 'loss/train': 0.8847519159317017}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:38 - INFO - __main__ - Step 140105: {'lr': 5.495511687872102e-06, 'samples': 26900160, 'steps': 140104, 'loss/train': 1.5512582063674927}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:39 - INFO - __main__ - Step 140109: {'lr': 5.491086305012493e-06, 'samples': 26900928, 'steps': 140108, 'loss/train': 1.5791300535202026}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:42 - INFO - __main__ - Step 140114: {'lr': 5.485557055310453e-06, 'samples': 26901888, 'steps': 140113, 'loss/train': 1.2267287969589233}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:42 - INFO - __main__ - Step 140114: {'lr': 5.485557055310453e-06, 'samples': 26901888, 'steps': 140113, 'loss/train': 1.2267287969589233}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:46 - INFO - __main__ - Step 140121: {'lr': 5.477820733067768e-06, 'samples': 26903232, 'steps': 140120, 'loss/train': 1.5690302848815918}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:47 - INFO - __main__ - Step 140125: {'lr': 5.473402401403815e-06, 'samples': 26904000, 'steps': 140124, 'loss/train': 1.3132870197296143}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:49 - INFO - __main__ - Step 140129: {'lr': 5.468985832618456e-06, 'samples': 26904768, 'steps': 140128, 'loss/train': 1.1601402759552002}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:52 - INFO - __main__ - Step 140134: {'lr': 5.4634676007332685e-06, 'samples': 26905728, 'steps': 140133, 'loss/train': 1.1840542554855347}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:54 - INFO - __main__ - Step 140139: {'lr': 5.457952123457788e-06, 'samples': 26906688, 'steps': 140138, 'loss/train': 1.0453351736068726}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:56 - INFO - __main__ - Step 140143: {'lr': 5.453541724998151e-06, 'samples': 26907456, 'steps': 140142, 'loss/train': 0.6671898365020752}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:56 - INFO - __main__ - Step 140143: {'lr': 5.453541724998151e-06, 'samples': 26907456, 'steps': 140142, 'loss/train': 0.6671898365020752}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:58:59 - INFO - __main__ - Step 140150: {'lr': 5.445827769984007e-06, 'samples': 26908800, 'steps': 140149, 'loss/train': 1.2330704927444458}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:02 - INFO - __main__ - Step 140155: {'lr': 5.440321107876928e-06, 'samples': 26909760, 'steps': 140154, 'loss/train': 1.6072440147399902}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:04 - INFO - __main__ - Step 140160: {'lr': 5.434817200640291e-06, 'samples': 26910720, 'steps': 140159, 'loss/train': 1.6604015827178955}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:04 - INFO - __main__ - Step 140160: {'lr': 5.434817200640291e-06, 'samples': 26910720, 'steps': 140159, 'loss/train': 1.6604015827178955}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:08 - INFO - __main__ - Step 140167: {'lr': 5.427116358809353e-06, 'samples': 26912064, 'steps': 140166, 'loss/train': 1.4106004238128662}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:09 - INFO - __main__ - Step 140171: {'lr': 5.4227183021690605e-06, 'samples': 26912832, 'steps': 140170, 'loss/train': 1.3396620750427246}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:09 - INFO - __main__ - Step 140171: {'lr': 5.4227183021690605e-06, 'samples': 26912832, 'steps': 140170, 'loss/train': 1.3396620750427246}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:09 - INFO - __main__ - Step 140171: {'lr': 5.4227183021690605e-06, 'samples': 26912832, 'steps': 140170, 'loss/train': 1.3396620750427246}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:15 - INFO - __main__ - Step 140182: {'lr': 5.4106327382304475e-06, 'samples': 26914944, 'steps': 140181, 'loss/train': 1.1012098789215088}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:18 - INFO - __main__ - Step 140187: {'lr': 5.405143708364702e-06, 'samples': 26915904, 'steps': 140186, 'loss/train': 1.1372592449188232}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:18 - INFO - __main__ - Step 140187: {'lr': 5.405143708364702e-06, 'samples': 26915904, 'steps': 140186, 'loss/train': 1.1372592449188232}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:22 - INFO - __main__ - Step 140195: {'lr': 5.396366991559987e-06, 'samples': 26917440, 'steps': 140194, 'loss/train': 1.7245769500732422}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:24 - INFO - __main__ - Step 140199: {'lr': 5.39198127829274e-06, 'samples': 26918208, 'steps': 140198, 'loss/train': 1.3046274185180664}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:24 - INFO - __main__ - Step 140199: {'lr': 5.39198127829274e-06, 'samples': 26918208, 'steps': 140198, 'loss/train': 1.3046274185180664}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:24 - INFO - __main__ - Step 140199: {'lr': 5.39198127829274e-06, 'samples': 26918208, 'steps': 140198, 'loss/train': 1.3046274185180664}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:30 - INFO - __main__ - Step 140210: {'lr': 5.3799296597722705e-06, 'samples': 26920320, 'steps': 140209, 'loss/train': 0.8647668361663818}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:32 - INFO - __main__ - Step 140216: {'lr': 5.373361670951704e-06, 'samples': 26921472, 'steps': 140215, 'loss/train': 1.0512995719909668}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:34 - INFO - __main__ - Step 140220: {'lr': 5.368985216231193e-06, 'samples': 26922240, 'steps': 140219, 'loss/train': 1.4387626647949219}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:37 - INFO - __main__ - Step 140224: {'lr': 5.364610525142033e-06, 'samples': 26923008, 'steps': 140223, 'loss/train': 1.523059368133545}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:39 - INFO - __main__ - Step 140228: {'lr': 5.360237597715811e-06, 'samples': 26923776, 'steps': 140227, 'loss/train': 0.5707634091377258}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:40 - INFO - __main__ - Step 140232: {'lr': 5.355866433983975e-06, 'samples': 26924544, 'steps': 140231, 'loss/train': 1.0319368839263916}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:42 - INFO - __main__ - Step 140236: {'lr': 5.351497033978137e-06, 'samples': 26925312, 'steps': 140235, 'loss/train': 1.4170215129852295}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:44 - INFO - __main__ - Step 140240: {'lr': 5.347129397729689e-06, 'samples': 26926080, 'steps': 140239, 'loss/train': 1.3455082178115845}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:44 - INFO - __main__ - Step 140240: {'lr': 5.347129397729689e-06, 'samples': 26926080, 'steps': 140239, 'loss/train': 1.3455082178115845}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:47 - INFO - __main__ - Step 140247: {'lr': 5.3394902784310025e-06, 'samples': 26927424, 'steps': 140246, 'loss/train': 1.6055904626846313}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:50 - INFO - __main__ - Step 140252: {'lr': 5.334037071843878e-06, 'samples': 26928384, 'steps': 140251, 'loss/train': 1.3426907062530518}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:50 - INFO - __main__ - Step 140252: {'lr': 5.334037071843878e-06, 'samples': 26928384, 'steps': 140251, 'loss/train': 1.3426907062530518}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:54 - INFO - __main__ - Step 140260: {'lr': 5.325317673950836e-06, 'samples': 26929920, 'steps': 140259, 'loss/train': 1.2476409673690796}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:56 - INFO - __main__ - Step 140264: {'lr': 5.320960620907866e-06, 'samples': 26930688, 'steps': 140263, 'loss/train': 1.188838243484497}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 16:59:58 - INFO - __main__ - Step 140268: {'lr': 5.3166053318425e-06, 'samples': 26931456, 'steps': 140267, 'loss/train': 1.5838003158569336}}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:00 - INFO - __main__ - Step 140273: {'lr': 5.311163701152089e-06, 'samples': 26932416, 'steps': 140272, 'loss/train': 1.470332145690918}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:00 - INFO - __main__ - Step 140273: {'lr': 5.311163701152089e-06, 'samples': 26932416, 'steps': 140272, 'loss/train': 1.470332145690918}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:04 - INFO - __main__ - Step 140281: {'lr': 5.302462825229642e-06, 'samples': 26933952, 'steps': 140280, 'loss/train': 1.4151508808135986}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:06 - INFO - __main__ - Step 140285: {'lr': 5.298115033419176e-06, 'samples': 26934720, 'steps': 140284, 'loss/train': 1.2908755540847778}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:08 - INFO - __main__ - Step 140289: {'lr': 5.293769005751014e-06, 'samples': 26935488, 'steps': 140288, 'loss/train': 1.4715303182601929}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:10 - INFO - __main__ - Step 140293: {'lr': 5.289424742256438e-06, 'samples': 26936256, 'steps': 140292, 'loss/train': 1.7879409790039062}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:12 - INFO - __main__ - Step 140298: {'lr': 5.283996893805126e-06, 'samples': 26937216, 'steps': 140297, 'loss/train': 1.3360569477081299}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:14 - INFO - __main__ - Step 140302: {'lr': 5.279656599815718e-06, 'samples': 26937984, 'steps': 140301, 'loss/train': 1.9711054563522339}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:14 - INFO - __main__ - Step 140302: {'lr': 5.279656599815718e-06, 'samples': 26937984, 'steps': 140301, 'loss/train': 1.9711054563522339}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:18 - INFO - __main__ - Step 140309: {'lr': 5.272065330640674e-06, 'samples': 26939328, 'steps': 140308, 'loss/train': 1.9964361190795898}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:20 - INFO - __main__ - Step 140314: {'lr': 5.266646303624778e-06, 'samples': 26940288, 'steps': 140313, 'loss/train': 0.9914337396621704}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:22 - INFO - __main__ - Step 140318: {'lr': 5.262313066924457e-06, 'samples': 26941056, 'steps': 140317, 'loss/train': 1.1633384227752686}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:24 - INFO - __main__ - Step 140322: {'lr': 5.257981594624539e-06, 'samples': 26941824, 'steps': 140321, 'loss/train': 1.0858997106552124}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:26 - INFO - __main__ - Step 140326: {'lr': 5.25365188675625e-06, 'samples': 26942592, 'steps': 140325, 'loss/train': 0.8618209958076477}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:28 - INFO - __main__ - Step 140330: {'lr': 5.249323943350815e-06, 'samples': 26943360, 'steps': 140329, 'loss/train': 0.8752583265304565}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:30 - INFO - __main__ - Step 140334: {'lr': 5.244997764439402e-06, 'samples': 26944128, 'steps': 140333, 'loss/train': 0.905376672744751}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:33 - INFO - __main__ - Step 140340: {'lr': 5.238511804566831e-06, 'samples': 26945280, 'steps': 140339, 'loss/train': 1.3389921188354492}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:35 - INFO - __main__ - Step 140344: {'lr': 5.234190037026992e-06, 'samples': 26946048, 'steps': 140343, 'loss/train': 1.0708869695663452}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:35 - INFO - __main__ - Step 140344: {'lr': 5.234190037026992e-06, 'samples': 26946048, 'steps': 140343, 'loss/train': 1.0708869695663452}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:38 - INFO - __main__ - Step 140351: {'lr': 5.226631189927278e-06, 'samples': 26947392, 'steps': 140350, 'loss/train': 1.2590469121932983}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:38 - INFO - __main__ - Step 140351: {'lr': 5.226631189927278e-06, 'samples': 26947392, 'steps': 140350, 'loss/train': 1.2590469121932983}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:42 - INFO - __main__ - Step 140357: {'lr': 5.220156479474242e-06, 'samples': 26948544, 'steps': 140356, 'loss/train': 0.46744075417518616}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:44 - INFO - __main__ - Step 140362: {'lr': 5.214763920511123e-06, 'samples': 26949504, 'steps': 140361, 'loss/train': 0.6512784957885742}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:44 - INFO - __main__ - Step 140362: {'lr': 5.214763920511123e-06, 'samples': 26949504, 'steps': 140361, 'loss/train': 0.6512784957885742}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:48 - INFO - __main__ - Step 140369: {'lr': 5.207218970427252e-06, 'samples': 26950848, 'steps': 140368, 'loss/train': 1.455683708190918}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:50 - INFO - __main__ - Step 140373: {'lr': 5.202909996965349e-06, 'samples': 26951616, 'steps': 140372, 'loss/train': 1.157201886177063}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:52 - INFO - __main__ - Step 140378: {'lr': 5.197526261931901e-06, 'samples': 26952576, 'steps': 140377, 'loss/train': 1.2605106830596924}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:55 - INFO - __main__ - Step 140382: {'lr': 5.193221259378156e-06, 'samples': 26953344, 'steps': 140381, 'loss/train': 0.936513364315033}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:57 - INFO - __main__ - Step 140386: {'lr': 5.188918021722888e-06, 'samples': 26954112, 'steps': 140385, 'loss/train': 1.8183033466339111}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:00:58 - INFO - __main__ - Step 140390: {'lr': 5.1846165489971846e-06, 'samples': 26954880, 'steps': 140389, 'loss/train': 1.334754467010498}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:00 - INFO - __main__ - Step 140394: {'lr': 5.180316841231991e-06, 'samples': 26955648, 'steps': 140393, 'loss/train': 1.4441243410110474}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:00 - INFO - __main__ - Step 140394: {'lr': 5.180316841231991e-06, 'samples': 26955648, 'steps': 140393, 'loss/train': 1.4441243410110474}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:04 - INFO - __main__ - Step 140402: {'lr': 5.171722720707262e-06, 'samples': 26957184, 'steps': 140401, 'loss/train': 1.5044152736663818}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:06 - INFO - __main__ - Step 140406: {'lr': 5.167428308009647e-06, 'samples': 26957952, 'steps': 140405, 'loss/train': 1.066625714302063}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:08 - INFO - __main__ - Step 140410: {'lr': 5.163135660396528e-06, 'samples': 26958720, 'steps': 140409, 'loss/train': 1.3654634952545166}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:10 - INFO - __main__ - Step 140415: {'lr': 5.157772333077265e-06, 'samples': 26959680, 'steps': 140414, 'loss/train': 1.3128982782363892}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:12 - INFO - __main__ - Step 140419: {'lr': 5.15348365701726e-06, 'samples': 26960448, 'steps': 140418, 'loss/train': 1.1609631776809692}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:15 - INFO - __main__ - Step 140423: {'lr': 5.14919674614231e-06, 'samples': 26961216, 'steps': 140422, 'loss/train': 1.6031166315078735}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:16 - INFO - __main__ - Step 140427: {'lr': 5.144911600483221e-06, 'samples': 26961984, 'steps': 140426, 'loss/train': 1.1052782535552979}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:18 - INFO - __main__ - Step 140431: {'lr': 5.140628220070914e-06, 'samples': 26962752, 'steps': 140430, 'loss/train': 1.618958830833435}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:18 - INFO - __main__ - Step 140431: {'lr': 5.140628220070914e-06, 'samples': 26962752, 'steps': 140430, 'loss/train': 1.618958830833435}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:23 - INFO - __main__ - Step 140439: {'lr': 5.132066755110215e-06, 'samples': 26964288, 'steps': 140438, 'loss/train': 1.2864205837249756}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:25 - INFO - __main__ - Step 140443: {'lr': 5.127788670623496e-06, 'samples': 26965056, 'steps': 140442, 'loss/train': 1.3600249290466309}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:26 - INFO - __main__ - Step 140447: {'lr': 5.123512351507042e-06, 'samples': 26965824, 'steps': 140446, 'loss/train': 1.5643372535705566}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:28 - INFO - __main__ - Step 140451: {'lr': 5.119237797791665e-06, 'samples': 26966592, 'steps': 140450, 'loss/train': 1.1445413827896118}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:31 - INFO - __main__ - Step 140457: {'lr': 5.112829277412967e-06, 'samples': 26967744, 'steps': 140456, 'loss/train': 1.3554588556289673}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:33 - INFO - __main__ - Step 140461: {'lr': 5.108559137335045e-06, 'samples': 26968512, 'steps': 140460, 'loss/train': 1.3161382675170898}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:33 - INFO - __main__ - Step 140461: {'lr': 5.108559137335045e-06, 'samples': 26968512, 'steps': 140460, 'loss/train': 1.3161382675170898}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:37 - INFO - __main__ - Step 140468: {'lr': 5.1010906404728994e-06, 'samples': 26969856, 'steps': 140467, 'loss/train': 1.1682419776916504}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:39 - INFO - __main__ - Step 140472: {'lr': 5.096825355618395e-06, 'samples': 26970624, 'steps': 140471, 'loss/train': 1.114585518836975}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:41 - INFO - __main__ - Step 140476: {'lr': 5.09256183635734e-06, 'samples': 26971392, 'steps': 140475, 'loss/train': 1.210998296737671}}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:43 - INFO - __main__ - Step 140480: {'lr': 5.088300082720487e-06, 'samples': 26972160, 'steps': 140479, 'loss/train': 1.079721450805664}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:44 - INFO - __main__ - Step 140484: {'lr': 5.08404009473859e-06, 'samples': 26972928, 'steps': 140483, 'loss/train': 0.7396302819252014}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:46 - INFO - __main__ - Step 140488: {'lr': 5.0797818724422905e-06, 'samples': 26973696, 'steps': 140487, 'loss/train': 1.8032175302505493}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:49 - INFO - __main__ - Step 140493: {'lr': 5.074461577614125e-06, 'samples': 26974656, 'steps': 140492, 'loss/train': 1.4346619844436646}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:51 - INFO - __main__ - Step 140497: {'lr': 5.070207328222748e-06, 'samples': 26975424, 'steps': 140496, 'loss/train': 1.2559436559677124}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:53 - INFO - __main__ - Step 140501: {'lr': 5.065954844616722e-06, 'samples': 26976192, 'steps': 140500, 'loss/train': 0.933594286441803}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:54 - INFO - __main__ - Step 140505: {'lr': 5.061704126826744e-06, 'samples': 26976960, 'steps': 140504, 'loss/train': 1.3695001602172852}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:56 - INFO - __main__ - Step 140509: {'lr': 5.0574551748833746e-06, 'samples': 26977728, 'steps': 140508, 'loss/train': 1.3501077890396118}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:01:59 - INFO - __main__ - Step 140514: {'lr': 5.052146468222746e-06, 'samples': 26978688, 'steps': 140513, 'loss/train': 1.1783499717712402}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:01 - INFO - __main__ - Step 140518: {'lr': 5.0479014895463695e-06, 'samples': 26979456, 'steps': 140517, 'loss/train': 1.3579448461532593}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:01 - INFO - __main__ - Step 140518: {'lr': 5.0479014895463695e-06, 'samples': 26979456, 'steps': 140517, 'loss/train': 1.3579448461532593}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:04 - INFO - __main__ - Step 140525: {'lr': 5.040477026189094e-06, 'samples': 26980800, 'steps': 140524, 'loss/train': 1.0752302408218384}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:06 - INFO - __main__ - Step 140530: {'lr': 5.035177149316644e-06, 'samples': 26981760, 'steps': 140529, 'loss/train': 1.3251954317092896}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:09 - INFO - __main__ - Step 140535: {'lr': 5.029880031878404e-06, 'samples': 26982720, 'steps': 140534, 'loss/train': 1.3282461166381836}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:09 - INFO - __main__ - Step 140535: {'lr': 5.029880031878404e-06, 'samples': 26982720, 'steps': 140534, 'loss/train': 1.3282461166381836}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:13 - INFO - __main__ - Step 140542: {'lr': 5.022468703427957e-06, 'samples': 26984064, 'steps': 140541, 'loss/train': 1.5225698947906494}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:14 - INFO - __main__ - Step 140546: {'lr': 5.018236087016764e-06, 'samples': 26984832, 'steps': 140545, 'loss/train': 1.1300032138824463}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:16 - INFO - __main__ - Step 140550: {'lr': 5.0140052367655676e-06, 'samples': 26985600, 'steps': 140549, 'loss/train': 1.6064437627792358}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:19 - INFO - __main__ - Step 140555: {'lr': 5.00871915766063e-06, 'samples': 26986560, 'steps': 140554, 'loss/train': 1.3606996536254883}8}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:21 - INFO - __main__ - Step 140560: {'lr': 5.0034358382879395e-06, 'samples': 26987520, 'steps': 140559, 'loss/train': 1.0809177160263062}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:23 - INFO - __main__ - Step 140564: {'lr': 4.999211169837037e-06, 'samples': 26988288, 'steps': 140563, 'loss/train': 1.0255731344223022}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:23 - INFO - __main__ - Step 140564: {'lr': 4.999211169837037e-06, 'samples': 26988288, 'steps': 140563, 'loss/train': 1.0255731344223022}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:26 - INFO - __main__ - Step 140571: {'lr': 4.99182225021888e-06, 'samples': 26989632, 'steps': 140570, 'loss/train': 1.225793480873108}2}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:29 - INFO - __main__ - Step 140576: {'lr': 4.986547762389215e-06, 'samples': 26990592, 'steps': 140575, 'loss/train': 1.2995396852493286}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:31 - INFO - __main__ - Step 140581: {'lr': 4.981276034541571e-06, 'samples': 26991552, 'steps': 140580, 'loss/train': 1.3643031120300293}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:31 - INFO - __main__ - Step 140581: {'lr': 4.981276034541571e-06, 'samples': 26991552, 'steps': 140580, 'loss/train': 1.3643031120300293}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:35 - INFO - __main__ - Step 140588: {'lr': 4.973900252437768e-06, 'samples': 26992896, 'steps': 140587, 'loss/train': 1.1666375398635864}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:37 - INFO - __main__ - Step 140592: {'lr': 4.969687948706087e-06, 'samples': 26993664, 'steps': 140591, 'loss/train': 1.1278413534164429}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:39 - INFO - __main__ - Step 140597: {'lr': 4.964425053199828e-06, 'samples': 26994624, 'steps': 140596, 'loss/train': 0.9435338973999023}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:39 - INFO - __main__ - Step 140597: {'lr': 4.964425053199828e-06, 'samples': 26994624, 'steps': 140596, 'loss/train': 0.9435338973999023}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:43 - INFO - __main__ - Step 140604: {'lr': 4.957061636692767e-06, 'samples': 26995968, 'steps': 140603, 'loss/train': 0.24603895843029022}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:45 - INFO - __main__ - Step 140608: {'lr': 4.9528563991833585e-06, 'samples': 26996736, 'steps': 140607, 'loss/train': 0.6173700094223022}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:47 - INFO - __main__ - Step 140613: {'lr': 4.947602336625529e-06, 'samples': 26997696, 'steps': 140612, 'loss/train': 1.5033385753631592}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:49 - INFO - __main__ - Step 140617: {'lr': 4.94340107407934e-06, 'samples': 26998464, 'steps': 140616, 'loss/train': 1.66404390335083}92}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:51 - INFO - __main__ - Step 140621: {'lr': 4.939201578232716e-06, 'samples': 26999232, 'steps': 140620, 'loss/train': 0.7965039610862732}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:53 - INFO - __main__ - Step 140625: {'lr': 4.93500384911591e-06, 'samples': 27000000, 'steps': 140624, 'loss/train': 1.5294981002807617}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:55 - INFO - __main__ - Step 140629: {'lr': 4.930807886759176e-06, 'samples': 27000768, 'steps': 140628, 'loss/train': 1.2352306842803955}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:02:57 - INFO - __main__ - Step 140634: {'lr': 4.925565418365752e-06, 'samples': 27001728, 'steps': 140633, 'loss/train': 0.9452007412910461}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:00 - INFO - __main__ - Step 140638: {'lr': 4.921373431329812e-06, 'samples': 27002496, 'steps': 140637, 'loss/train': 1.3364495038986206}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:00 - INFO - __main__ - Step 140638: {'lr': 4.921373431329812e-06, 'samples': 27002496, 'steps': 140637, 'loss/train': 1.3364495038986206}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:03 - INFO - __main__ - Step 140645: {'lr': 4.914041705538025e-06, 'samples': 27003840, 'steps': 140644, 'loss/train': 1.418485164642334}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:05 - INFO - __main__ - Step 140650: {'lr': 4.908808071492965e-06, 'samples': 27004800, 'steps': 140649, 'loss/train': 1.549443006515503}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:08 - INFO - __main__ - Step 140654: {'lr': 4.904623152071686e-06, 'samples': 27005568, 'steps': 140653, 'loss/train': 1.5700104236602783}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:08 - INFO - __main__ - Step 140654: {'lr': 4.904623152071686e-06, 'samples': 27005568, 'steps': 140653, 'loss/train': 1.5700104236602783}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:11 - INFO - __main__ - Step 140661: {'lr': 4.89730379489603e-06, 'samples': 27006912, 'steps': 140660, 'loss/train': 1.1419135332107544}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:13 - INFO - __main__ - Step 140665: {'lr': 4.893123734739852e-06, 'samples': 27007680, 'steps': 140664, 'loss/train': 1.033457636833191}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:13 - INFO - __main__ - Step 140665: {'lr': 4.893123734739852e-06, 'samples': 27007680, 'steps': 140664, 'loss/train': 1.033457636833191}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:17 - INFO - __main__ - Step 140673: {'lr': 4.8847689156436555e-06, 'samples': 27009216, 'steps': 140672, 'loss/train': 1.2839852571487427}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:17 - INFO - __main__ - Step 140673: {'lr': 4.8847689156436555e-06, 'samples': 27009216, 'steps': 140672, 'loss/train': 1.2839852571487427}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:21 - INFO - __main__ - Step 140680: {'lr': 4.877464247296154e-06, 'samples': 27010560, 'steps': 140679, 'loss/train': 1.358144998550415}7}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:23 - INFO - __main__ - Step 140685: {'lr': 4.872249940491486e-06, 'samples': 27011520, 'steps': 140684, 'loss/train': 1.216846227645874}7}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:23 - INFO - __main__ - Step 140685: {'lr': 4.872249940491486e-06, 'samples': 27011520, 'steps': 140684, 'loss/train': 1.216846227645874}7}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:27 - INFO - __main__ - Step 140692: {'lr': 4.8649545499109545e-06, 'samples': 27012864, 'steps': 140691, 'loss/train': 1.5745768547058105}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:29 - INFO - __main__ - Step 140696: {'lr': 4.860788185272663e-06, 'samples': 27013632, 'steps': 140695, 'loss/train': 1.1481571197509766}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:31 - INFO - __main__ - Step 140701: {'lr': 4.855582714737128e-06, 'samples': 27014592, 'steps': 140700, 'loss/train': 1.261768102645874}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:31 - INFO - __main__ - Step 140701: {'lr': 4.855582714737128e-06, 'samples': 27014592, 'steps': 140700, 'loss/train': 1.261768102645874}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:31 - INFO - __main__ - Step 140701: {'lr': 4.855582714737128e-06, 'samples': 27014592, 'steps': 140700, 'loss/train': 1.261768102645874}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:37 - INFO - __main__ - Step 140712: {'lr': 4.8441403999711085e-06, 'samples': 27016704, 'steps': 140711, 'loss/train': 1.0562560558319092}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:39 - INFO - __main__ - Step 140718: {'lr': 4.837904770967461e-06, 'samples': 27017856, 'steps': 140717, 'loss/train': 1.5212595462799072}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:42 - INFO - __main__ - Step 140722: {'lr': 4.833749894283896e-06, 'samples': 27018624, 'steps': 140721, 'loss/train': 1.340980887413025}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:42 - INFO - __main__ - Step 140722: {'lr': 4.833749894283896e-06, 'samples': 27018624, 'steps': 140721, 'loss/train': 1.340980887413025}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:45 - INFO - __main__ - Step 140729: {'lr': 4.8264831131286655e-06, 'samples': 27019968, 'steps': 140728, 'loss/train': 1.0828857421875}}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:47 - INFO - __main__ - Step 140733: {'lr': 4.822333097114856e-06, 'samples': 27020736, 'steps': 140732, 'loss/train': 1.6933735609054565}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:50 - INFO - __main__ - Step 140739: {'lr': 4.816111387301647e-06, 'samples': 27021888, 'steps': 140738, 'loss/train': 1.4657477140426636}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:50 - INFO - __main__ - Step 140739: {'lr': 4.816111387301647e-06, 'samples': 27021888, 'steps': 140738, 'loss/train': 1.4657477140426636}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:53 - INFO - __main__ - Step 140746: {'lr': 4.808857752521489e-06, 'samples': 27023232, 'steps': 140745, 'loss/train': 1.8173776865005493}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:55 - INFO - __main__ - Step 140750: {'lr': 4.8047152488967235e-06, 'samples': 27024000, 'steps': 140749, 'loss/train': 1.149765968322754}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:57 - INFO - __main__ - Step 140754: {'lr': 4.800574512971334e-06, 'samples': 27024768, 'steps': 140753, 'loss/train': 1.164465069770813}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:03:59 - INFO - __main__ - Step 140759: {'lr': 4.795401078937334e-06, 'samples': 27025728, 'steps': 140758, 'loss/train': 1.659677505493164}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:02 - INFO - __main__ - Step 140764: {'lr': 4.790230407038576e-06, 'samples': 27026688, 'steps': 140763, 'loss/train': 1.5065218210220337}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:02 - INFO - __main__ - Step 140764: {'lr': 4.790230407038576e-06, 'samples': 27026688, 'steps': 140763, 'loss/train': 1.5065218210220337}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:05 - INFO - __main__ - Step 140771: {'lr': 4.782996106878323e-06, 'samples': 27028032, 'steps': 140770, 'loss/train': 1.3799818754196167}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:07 - INFO - __main__ - Step 140775: {'lr': 4.778864651864195e-06, 'samples': 27028800, 'steps': 140774, 'loss/train': 1.1081119775772095}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:10 - INFO - __main__ - Step 140780: {'lr': 4.7737028191893465e-06, 'samples': 27029760, 'steps': 140779, 'loss/train': 1.30923593044281}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:12 - INFO - __main__ - Step 140785: {'lr': 4.768543748894155e-06, 'samples': 27030720, 'steps': 140784, 'loss/train': 1.915226936340332}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:12 - INFO - __main__ - Step 140785: {'lr': 4.768543748894155e-06, 'samples': 27030720, 'steps': 140784, 'loss/train': 1.915226936340332}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:15 - INFO - __main__ - Step 140792: {'lr': 4.761325691389262e-06, 'samples': 27032064, 'steps': 140791, 'loss/train': 1.210250735282898}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:17 - INFO - __main__ - Step 140796: {'lr': 4.757203518107323e-06, 'samples': 27032832, 'steps': 140795, 'loss/train': 1.3440746068954468}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:20 - INFO - __main__ - Step 140801: {'lr': 4.752053287817332e-06, 'samples': 27033792, 'steps': 140800, 'loss/train': 1.0921456813812256}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:22 - INFO - __main__ - Step 140805: {'lr': 4.747935092671435e-06, 'samples': 27034560, 'steps': 140804, 'loss/train': 1.4294623136520386}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:24 - INFO - __main__ - Step 140809: {'lr': 4.743818665634309e-06, 'samples': 27035328, 'steps': 140808, 'loss/train': 1.6042720079421997}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:26 - INFO - __main__ - Step 140813: {'lr': 4.739704006735596e-06, 'samples': 27036096, 'steps': 140812, 'loss/train': 1.3080891370773315}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:27 - INFO - __main__ - Step 140817: {'lr': 4.7355911160049666e-06, 'samples': 27036864, 'steps': 140816, 'loss/train': 1.0067697763442993}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:27 - INFO - __main__ - Step 140817: {'lr': 4.7355911160049666e-06, 'samples': 27036864, 'steps': 140816, 'loss/train': 1.0067697763442993}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:27 - INFO - __main__ - Step 140817: {'lr': 4.7355911160049666e-06, 'samples': 27036864, 'steps': 140816, 'loss/train': 1.0067697763442993}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:33 - INFO - __main__ - Step 140828: {'lr': 4.724289783854363e-06, 'samples': 27038976, 'steps': 140827, 'loss/train': 1.0480762720108032}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:35 - INFO - __main__ - Step 140833: {'lr': 4.7191572353559584e-06, 'samples': 27039936, 'steps': 140832, 'loss/train': 1.1929739713668823}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:38 - INFO - __main__ - Step 140838: {'lr': 4.7140274498516375e-06, 'samples': 27040896, 'steps': 140837, 'loss/train': 1.422528624534607}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:40 - INFO - __main__ - Step 140843: {'lr': 4.708900427399188e-06, 'samples': 27041856, 'steps': 140842, 'loss/train': 1.421828031539917}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:40 - INFO - __main__ - Step 140843: {'lr': 4.708900427399188e-06, 'samples': 27041856, 'steps': 140842, 'loss/train': 1.421828031539917}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:43 - INFO - __main__ - Step 140849: {'lr': 4.702751647765985e-06, 'samples': 27043008, 'steps': 140848, 'loss/train': 1.4317413568496704}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:45 - INFO - __main__ - Step 140853: {'lr': 4.698654671880825e-06, 'samples': 27043776, 'steps': 140852, 'loss/train': 1.7229357957839966}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:48 - INFO - __main__ - Step 140858: {'lr': 4.6935359389303214e-06, 'samples': 27044736, 'steps': 140857, 'loss/train': 1.6513245105743408}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:48 - INFO - __main__ - Step 140858: {'lr': 4.6935359389303214e-06, 'samples': 27044736, 'steps': 140857, 'loss/train': 1.6513245105743408}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:51 - INFO - __main__ - Step 140865: {'lr': 4.686374355127421e-06, 'samples': 27046080, 'steps': 140864, 'loss/train': 1.705886960029602}8}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:53 - INFO - __main__ - Step 140869: {'lr': 4.682284453275399e-06, 'samples': 27046848, 'steps': 140868, 'loss/train': 0.9406264424324036}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:53 - INFO - __main__ - Step 140869: {'lr': 4.682284453275399e-06, 'samples': 27046848, 'steps': 140868, 'loss/train': 0.9406264424324036}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:58 - INFO - __main__ - Step 140878: {'lr': 4.673088640531259e-06, 'samples': 27048576, 'steps': 140877, 'loss/train': 0.9533061981201172}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:04:58 - INFO - __main__ - Step 140878: {'lr': 4.673088640531259e-06, 'samples': 27048576, 'steps': 140877, 'loss/train': 0.9533061981201172}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:01 - INFO - __main__ - Step 140885: {'lr': 4.665942531982242e-06, 'samples': 27049920, 'steps': 140884, 'loss/train': 1.6230487823486328}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:03 - INFO - __main__ - Step 140890: {'lr': 4.660841485035955e-06, 'samples': 27050880, 'steps': 140889, 'loss/train': 1.39169180393219}8}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:03 - INFO - __main__ - Step 140890: {'lr': 4.660841485035955e-06, 'samples': 27050880, 'steps': 140889, 'loss/train': 1.39169180393219}8}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:08 - INFO - __main__ - Step 140898: {'lr': 4.652685558339808e-06, 'samples': 27052416, 'steps': 140897, 'loss/train': 1.22709059715271}8}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:09 - INFO - __main__ - Step 140902: {'lr': 4.648610248170176e-06, 'samples': 27053184, 'steps': 140901, 'loss/train': 1.4857808351516724}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:11 - INFO - __main__ - Step 140906: {'lr': 4.644536706825353e-06, 'samples': 27053952, 'steps': 140905, 'loss/train': 1.6522880792617798}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:14 - INFO - __main__ - Step 140911: {'lr': 4.639447267598934e-06, 'samples': 27054912, 'steps': 140910, 'loss/train': 1.412675142288208}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:16 - INFO - __main__ - Step 140916: {'lr': 4.634360592264463e-06, 'samples': 27055872, 'steps': 140915, 'loss/train': 1.486100435256958}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:16 - INFO - __main__ - Step 140916: {'lr': 4.634360592264463e-06, 'samples': 27055872, 'steps': 140915, 'loss/train': 1.486100435256958}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:19 - INFO - __main__ - Step 140923: {'lr': 4.627243890243854e-06, 'samples': 27057216, 'steps': 140922, 'loss/train': 1.5800294876098633}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:21 - INFO - __main__ - Step 140927: {'lr': 4.623179635710739e-06, 'samples': 27057984, 'steps': 140926, 'loss/train': 1.371856927871704}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:24 - INFO - __main__ - Step 140932: {'lr': 4.618101805215491e-06, 'samples': 27058944, 'steps': 140931, 'loss/train': 1.2919220924377441}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:26 - INFO - __main__ - Step 140936: {'lr': 4.6140415309919026e-06, 'samples': 27059712, 'steps': 140935, 'loss/train': 1.2444671392440796}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:28 - INFO - __main__ - Step 140940: {'lr': 4.609983025842313e-06, 'samples': 27060480, 'steps': 140939, 'loss/train': 1.0811411142349243}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:28 - INFO - __main__ - Step 140940: {'lr': 4.609983025842313e-06, 'samples': 27060480, 'steps': 140939, 'loss/train': 1.0811411142349243}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:31 - INFO - __main__ - Step 140947: {'lr': 4.602884898752907e-06, 'samples': 27061824, 'steps': 140946, 'loss/train': 0.3413761854171753}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:34 - INFO - __main__ - Step 140952: {'lr': 4.597818125130215e-06, 'samples': 27062784, 'steps': 140951, 'loss/train': 0.9250797629356384}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:34 - INFO - __main__ - Step 140952: {'lr': 4.597818125130215e-06, 'samples': 27062784, 'steps': 140951, 'loss/train': 0.9250797629356384}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:38 - INFO - __main__ - Step 140960: {'lr': 4.589717037228353e-06, 'samples': 27064320, 'steps': 140959, 'loss/train': 1.1385483741760254}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:40 - INFO - __main__ - Step 140964: {'lr': 4.585669147136873e-06, 'samples': 27065088, 'steps': 140963, 'loss/train': 0.7734197378158569}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:42 - INFO - __main__ - Step 140968: {'lr': 4.581623026323978e-06, 'samples': 27065856, 'steps': 140967, 'loss/train': 1.38002347946167}9}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:44 - INFO - __main__ - Step 140973: {'lr': 4.5765678634003e-06, 'samples': 27066816, 'steps': 140972, 'loss/train': 1.3533074855804443}9}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:46 - INFO - __main__ - Step 140977: {'lr': 4.572525723570809e-06, 'samples': 27067584, 'steps': 140976, 'loss/train': 1.5227785110473633}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:46 - INFO - __main__ - Step 140977: {'lr': 4.572525723570809e-06, 'samples': 27067584, 'steps': 140976, 'loss/train': 1.5227785110473633}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:50 - INFO - __main__ - Step 140984: {'lr': 4.5654562364412786e-06, 'samples': 27068928, 'steps': 140983, 'loss/train': 1.2922940254211426}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:52 - INFO - __main__ - Step 140989: {'lr': 4.560409920438796e-06, 'samples': 27069888, 'steps': 140988, 'loss/train': 1.0380473136901855}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:54 - INFO - __main__ - Step 140994: {'lr': 4.555366369218578e-06, 'samples': 27070848, 'steps': 140993, 'loss/train': 0.673716127872467}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:56 - INFO - __main__ - Step 140998: {'lr': 4.551333518923867e-06, 'samples': 27071616, 'steps': 140997, 'loss/train': 1.1755448579788208}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:05:56 - INFO - __main__ - Step 140998: {'lr': 4.551333518923867e-06, 'samples': 27071616, 'steps': 140997, 'loss/train': 1.1755448579788208}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:00 - INFO - __main__ - Step 141005: {'lr': 4.5442802888477355e-06, 'samples': 27072960, 'steps': 141004, 'loss/train': 1.894527792930603}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:00 - INFO - __main__ - Step 141005: {'lr': 4.5442802888477355e-06, 'samples': 27072960, 'steps': 141004, 'loss/train': 1.894527792930603}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:04 - INFO - __main__ - Step 141013: {'lr': 4.536226090397694e-06, 'samples': 27074496, 'steps': 141012, 'loss/train': 1.0356035232543945}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:06 - INFO - __main__ - Step 141017: {'lr': 4.53220164561069e-06, 'samples': 27075264, 'steps': 141016, 'loss/train': 1.026716947555542}5}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:08 - INFO - __main__ - Step 141021: {'lr': 4.52817897048774e-06, 'samples': 27076032, 'steps': 141020, 'loss/train': 1.1017184257507324}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:10 - INFO - __main__ - Step 141026: {'lr': 4.523153115218226e-06, 'samples': 27076992, 'steps': 141025, 'loss/train': 1.276179313659668}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:12 - INFO - __main__ - Step 141030: {'lr': 4.519134421945348e-06, 'samples': 27077760, 'steps': 141029, 'loss/train': 1.4851237535476685}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:12 - INFO - __main__ - Step 141030: {'lr': 4.519134421945348e-06, 'samples': 27077760, 'steps': 141029, 'loss/train': 1.4851237535476685}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:15 - INFO - __main__ - Step 141036: {'lr': 4.513109700341772e-06, 'samples': 27078912, 'steps': 141035, 'loss/train': 1.1147538423538208}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:18 - INFO - __main__ - Step 141041: {'lr': 4.508092140847936e-06, 'samples': 27079872, 'steps': 141040, 'loss/train': 1.353209376335144}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:18 - INFO - __main__ - Step 141041: {'lr': 4.508092140847936e-06, 'samples': 27079872, 'steps': 141040, 'loss/train': 1.353209376335144}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:22 - INFO - __main__ - Step 141049: {'lr': 4.500069797654299e-06, 'samples': 27081408, 'steps': 141048, 'loss/train': 0.4233951270580292}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:23 - INFO - __main__ - Step 141053: {'lr': 4.49606128088656e-06, 'samples': 27082176, 'steps': 141052, 'loss/train': 1.6235346794128418}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:25 - INFO - __main__ - Step 141057: {'lr': 4.492054534043389e-06, 'samples': 27082944, 'steps': 141056, 'loss/train': 1.4769694805145264}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:28 - INFO - __main__ - Step 141062: {'lr': 4.4870485894898675e-06, 'samples': 27083904, 'steps': 141061, 'loss/train': 1.1366404294967651}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:28 - INFO - __main__ - Step 141062: {'lr': 4.4870485894898675e-06, 'samples': 27083904, 'steps': 141061, 'loss/train': 1.1366404294967651}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:32 - INFO - __main__ - Step 141070: {'lr': 4.479044830693779e-06, 'samples': 27085440, 'steps': 141069, 'loss/train': 1.4739705324172974}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:34 - INFO - __main__ - Step 141074: {'lr': 4.475045606352174e-06, 'samples': 27086208, 'steps': 141073, 'loss/train': 1.0093538761138916}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:36 - INFO - __main__ - Step 141078: {'lr': 4.471048152086682e-06, 'samples': 27086976, 'steps': 141077, 'loss/train': 1.0471833944320679}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:38 - INFO - __main__ - Step 141083: {'lr': 4.466053823468236e-06, 'samples': 27087936, 'steps': 141082, 'loss/train': 1.157873511314392}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:40 - INFO - __main__ - Step 141088: {'lr': 4.4610622606949535e-06, 'samples': 27088896, 'steps': 141087, 'loss/train': 1.587041974067688}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:40 - INFO - __main__ - Step 141088: {'lr': 4.4610622606949535e-06, 'samples': 27088896, 'steps': 141087, 'loss/train': 1.587041974067688}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:40 - INFO - __main__ - Step 141088: {'lr': 4.4610622606949535e-06, 'samples': 27088896, 'steps': 141087, 'loss/train': 1.587041974067688}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:46 - INFO - __main__ - Step 141098: {'lr': 4.451087432908813e-06, 'samples': 27090816, 'steps': 141097, 'loss/train': 1.4274747371673584}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:48 - INFO - __main__ - Step 141103: {'lr': 4.446104168008308e-06, 'samples': 27091776, 'steps': 141102, 'loss/train': 1.7361775636672974}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:50 - INFO - __main__ - Step 141107: {'lr': 4.4421195476555265e-06, 'samples': 27092544, 'steps': 141106, 'loss/train': 1.677897572517395}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:52 - INFO - __main__ - Step 141111: {'lr': 4.4381366976162516e-06, 'samples': 27093312, 'steps': 141110, 'loss/train': 1.3067580461502075}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:54 - INFO - __main__ - Step 141115: {'lr': 4.434155617919127e-06, 'samples': 27094080, 'steps': 141114, 'loss/train': 1.1571290493011475}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:56 - INFO - __main__ - Step 141119: {'lr': 4.430176308592909e-06, 'samples': 27094848, 'steps': 141118, 'loss/train': 1.3470758199691772}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:58 - INFO - __main__ - Step 141124: {'lr': 4.425204661562876e-06, 'samples': 27095808, 'steps': 141123, 'loss/train': 1.3405522108078003}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:06:58 - INFO - __main__ - Step 141124: {'lr': 4.425204661562876e-06, 'samples': 27095808, 'steps': 141123, 'loss/train': 1.3405522108078003}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:02 - INFO - __main__ - Step 141131: {'lr': 4.4182490031263175e-06, 'samples': 27097152, 'steps': 141130, 'loss/train': 1.014966607093811}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:04 - INFO - __main__ - Step 141135: {'lr': 4.4142767755703805e-06, 'samples': 27097920, 'steps': 141134, 'loss/train': 1.155704379081726}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:07 - INFO - __main__ - Step 141140: {'lr': 4.409313980914376e-06, 'samples': 27098880, 'steps': 141139, 'loss/train': 1.6564116477966309}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:07 - INFO - __main__ - Step 141140: {'lr': 4.409313980914376e-06, 'samples': 27098880, 'steps': 141139, 'loss/train': 1.6564116477966309}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:07 - INFO - __main__ - Step 141140: {'lr': 4.409313980914376e-06, 'samples': 27098880, 'steps': 141139, 'loss/train': 1.6564116477966309}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:12 - INFO - __main__ - Step 141151: {'lr': 4.3984055707747274e-06, 'samples': 27100992, 'steps': 141150, 'loss/train': 2.2499430179595947}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:14 - INFO - __main__ - Step 141156: {'lr': 4.393451629065209e-06, 'samples': 27101952, 'steps': 141155, 'loss/train': 1.1708078384399414}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:14 - INFO - __main__ - Step 141156: {'lr': 4.393451629065209e-06, 'samples': 27101952, 'steps': 141155, 'loss/train': 1.1708078384399414}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:18 - INFO - __main__ - Step 141163: {'lr': 4.386520758678292e-06, 'samples': 27103296, 'steps': 141162, 'loss/train': 1.3480803966522217}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:20 - INFO - __main__ - Step 141168: {'lr': 4.3815734570537005e-06, 'samples': 27104256, 'steps': 141167, 'loss/train': 0.9961921572685242}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:23 - INFO - __main__ - Step 141173: {'lr': 4.376628922225956e-06, 'samples': 27105216, 'steps': 141172, 'loss/train': 1.3895741701126099}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:23 - INFO - __main__ - Step 141173: {'lr': 4.376628922225956e-06, 'samples': 27105216, 'steps': 141172, 'loss/train': 1.3895741701126099}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:27 - INFO - __main__ - Step 141180: {'lr': 4.369711221791805e-06, 'samples': 27106560, 'steps': 141179, 'loss/train': 1.2672178745269775}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:28 - INFO - __main__ - Step 141184: {'lr': 4.365760685004161e-06, 'samples': 27107328, 'steps': 141183, 'loss/train': 1.3516994714736938}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:28 - INFO - __main__ - Step 141184: {'lr': 4.365760685004161e-06, 'samples': 27107328, 'steps': 141183, 'loss/train': 1.3516994714736938}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:28 - INFO - __main__ - Step 141184: {'lr': 4.365760685004161e-06, 'samples': 27107328, 'steps': 141183, 'loss/train': 1.3516994714736938}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:34 - INFO - __main__ - Step 141195: {'lr': 4.354905839941026e-06, 'samples': 27109440, 'steps': 141194, 'loss/train': 0.9511042833328247}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:36 - INFO - __main__ - Step 141199: {'lr': 4.350961944039972e-06, 'samples': 27110208, 'steps': 141198, 'loss/train': 1.2642667293548584}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:39 - INFO - __main__ - Step 141204: {'lr': 4.346034564594509e-06, 'samples': 27111168, 'steps': 141203, 'loss/train': 1.5046992301940918}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:41 - INFO - __main__ - Step 141208: {'lr': 4.342094653417394e-06, 'samples': 27111936, 'steps': 141207, 'loss/train': 1.5707738399505615}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:41 - INFO - __main__ - Step 141208: {'lr': 4.342094653417394e-06, 'samples': 27111936, 'steps': 141207, 'loss/train': 1.5707738399505615}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:44 - INFO - __main__ - Step 141216: {'lr': 4.334220144195323e-06, 'samples': 27113472, 'steps': 141215, 'loss/train': 1.2300747632980347}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:47 - INFO - __main__ - Step 141220: {'lr': 4.330285546207125e-06, 'samples': 27114240, 'steps': 141219, 'loss/train': 1.466942310333252}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:49 - INFO - __main__ - Step 141225: {'lr': 4.3253697893622935e-06, 'samples': 27115200, 'steps': 141224, 'loss/train': 1.359566569328308}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:49 - INFO - __main__ - Step 141225: {'lr': 4.3253697893622935e-06, 'samples': 27115200, 'steps': 141224, 'loss/train': 1.359566569328308}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:53 - INFO - __main__ - Step 141232: {'lr': 4.318492379074224e-06, 'samples': 27116544, 'steps': 141231, 'loss/train': 1.1215802431106567}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:54 - INFO - __main__ - Step 141236: {'lr': 4.3145648657349765e-06, 'samples': 27117312, 'steps': 141235, 'loss/train': 1.347169041633606}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:56 - INFO - __main__ - Step 141240: {'lr': 4.310639123628751e-06, 'samples': 27118080, 'steps': 141239, 'loss/train': 0.7246345281600952}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:59 - INFO - __main__ - Step 141245: {'lr': 4.305734436835601e-06, 'samples': 27119040, 'steps': 141244, 'loss/train': 1.1994625329971313}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:07:59 - INFO - __main__ - Step 141245: {'lr': 4.305734436835601e-06, 'samples': 27119040, 'steps': 141244, 'loss/train': 1.1994625329971313}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:02 - INFO - __main__ - Step 141251: {'lr': 4.29985246598813e-06, 'samples': 27120192, 'steps': 141250, 'loss/train': 1.2963447570800781}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:05 - INFO - __main__ - Step 141256: {'lr': 4.294953868100332e-06, 'samples': 27121152, 'steps': 141255, 'loss/train': 1.2921475172042847}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:05 - INFO - __main__ - Step 141256: {'lr': 4.294953868100332e-06, 'samples': 27121152, 'steps': 141255, 'loss/train': 1.2921475172042847}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:08 - INFO - __main__ - Step 141263: {'lr': 4.28810048092762e-06, 'samples': 27122496, 'steps': 141262, 'loss/train': 0.6664243340492249}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:10 - INFO - __main__ - Step 141267: {'lr': 4.284186695384163e-06, 'samples': 27123264, 'steps': 141266, 'loss/train': 1.4787243604660034}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:13 - INFO - __main__ - Step 141272: {'lr': 4.279296954562612e-06, 'samples': 27124224, 'steps': 141271, 'loss/train': 1.0764625072479248}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:15 - INFO - __main__ - Step 141277: {'lr': 4.274409981689958e-06, 'samples': 27125184, 'steps': 141276, 'loss/train': 0.9479673504829407}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:15 - INFO - __main__ - Step 141277: {'lr': 4.274409981689958e-06, 'samples': 27125184, 'steps': 141276, 'loss/train': 0.9479673504829407}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:18 - INFO - __main__ - Step 141284: {'lr': 4.2675728699272946e-06, 'samples': 27126528, 'steps': 141283, 'loss/train': 1.1851589679718018}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:20 - INFO - __main__ - Step 141288: {'lr': 4.263668384821645e-06, 'samples': 27127296, 'steps': 141287, 'loss/train': 0.07607757300138474}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:23 - INFO - __main__ - Step 141293: {'lr': 4.2587902697553695e-06, 'samples': 27128256, 'steps': 141292, 'loss/train': 1.2350516319274902}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:23 - INFO - __main__ - Step 141293: {'lr': 4.2587902697553695e-06, 'samples': 27128256, 'steps': 141292, 'loss/train': 1.2350516319274902}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:27 - INFO - __main__ - Step 141300: {'lr': 4.251965559217141e-06, 'samples': 27129600, 'steps': 141299, 'loss/train': 1.3758069276809692}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:28 - INFO - __main__ - Step 141304: {'lr': 4.248068160680185e-06, 'samples': 27130368, 'steps': 141303, 'loss/train': 1.5156867504119873}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:31 - INFO - __main__ - Step 141308: {'lr': 4.244172533855673e-06, 'samples': 27131136, 'steps': 141307, 'loss/train': 1.5576553344726562}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:33 - INFO - __main__ - Step 141313: {'lr': 4.2393054918384855e-06, 'samples': 27132096, 'steps': 141312, 'loss/train': 1.301392912864685}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:35 - INFO - __main__ - Step 141317: {'lr': 4.235413851469666e-06, 'samples': 27132864, 'steps': 141316, 'loss/train': 1.2836394309997559}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:37 - INFO - __main__ - Step 141321: {'lr': 4.231523982904523e-06, 'samples': 27133632, 'steps': 141320, 'loss/train': 1.6494863033294678}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:39 - INFO - __main__ - Step 141325: {'lr': 4.227635886171116e-06, 'samples': 27134400, 'steps': 141324, 'loss/train': 1.1892045736312866}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:41 - INFO - __main__ - Step 141329: {'lr': 4.223749561297452e-06, 'samples': 27135168, 'steps': 141328, 'loss/train': 1.284403920173645}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:43 - INFO - __main__ - Step 141334: {'lr': 4.218894146925833e-06, 'samples': 27136128, 'steps': 141333, 'loss/train': 1.4444152116775513}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:45 - INFO - __main__ - Step 141338: {'lr': 4.21501180883907e-06, 'samples': 27136896, 'steps': 141337, 'loss/train': 0.5214104056358337}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:45 - INFO - __main__ - Step 141338: {'lr': 4.21501180883907e-06, 'samples': 27136896, 'steps': 141337, 'loss/train': 0.5214104056358337}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:49 - INFO - __main__ - Step 141345: {'lr': 4.208221980960547e-06, 'samples': 27138240, 'steps': 141344, 'loss/train': 1.5674155950546265}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:51 - INFO - __main__ - Step 141350: {'lr': 4.203375426395206e-06, 'samples': 27139200, 'steps': 141349, 'loss/train': 1.5487823486328125}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:51 - INFO - __main__ - Step 141350: {'lr': 4.203375426395206e-06, 'samples': 27139200, 'steps': 141349, 'loss/train': 1.5487823486328125}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:55 - INFO - __main__ - Step 141357: {'lr': 4.196594901606898e-06, 'samples': 27140544, 'steps': 141356, 'loss/train': 0.6116508841514587}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:56 - INFO - __main__ - Step 141361: {'lr': 4.192722752618866e-06, 'samples': 27141312, 'steps': 141360, 'loss/train': 0.8957082629203796}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:59 - INFO - __main__ - Step 141366: {'lr': 4.187885058418828e-06, 'samples': 27142272, 'steps': 141365, 'loss/train': 0.8011024594306946}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:08:59 - INFO - __main__ - Step 141366: {'lr': 4.187885058418828e-06, 'samples': 27142272, 'steps': 141365, 'loss/train': 0.8011024594306946}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:03 - INFO - __main__ - Step 141374: {'lr': 4.180150507196973e-06, 'samples': 27143808, 'steps': 141373, 'loss/train': 0.890661358833313}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:03 - INFO - __main__ - Step 141374: {'lr': 4.180150507196973e-06, 'samples': 27143808, 'steps': 141373, 'loss/train': 0.890661358833313}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:06 - INFO - __main__ - Step 141381: {'lr': 4.173388589908372e-06, 'samples': 27145152, 'steps': 141380, 'loss/train': 1.6019949913024902}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:09 - INFO - __main__ - Step 141386: {'lr': 4.168561971949536e-06, 'samples': 27146112, 'steps': 141385, 'loss/train': 1.3558725118637085}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:09 - INFO - __main__ - Step 141386: {'lr': 4.168561971949536e-06, 'samples': 27146112, 'steps': 141385, 'loss/train': 1.3558725118637085}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:13 - INFO - __main__ - Step 141394: {'lr': 4.160845143166392e-06, 'samples': 27147648, 'steps': 141393, 'loss/train': 0.9315038919448853}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:14 - INFO - __main__ - Step 141398: {'lr': 4.156989387272797e-06, 'samples': 27148416, 'steps': 141397, 'loss/train': 1.727115273475647}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:16 - INFO - __main__ - Step 141402: {'lr': 4.153135403748287e-06, 'samples': 27149184, 'steps': 141401, 'loss/train': 1.2841302156448364}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:19 - INFO - __main__ - Step 141407: {'lr': 4.148320416779011e-06, 'samples': 27150144, 'steps': 141406, 'loss/train': 1.5303804874420166}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:19 - INFO - __main__ - Step 141407: {'lr': 4.148320416779011e-06, 'samples': 27150144, 'steps': 141406, 'loss/train': 1.5303804874420166}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:23 - INFO - __main__ - Step 141415: {'lr': 4.140622198053251e-06, 'samples': 27151680, 'steps': 141414, 'loss/train': 1.3930096626281738}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:25 - INFO - __main__ - Step 141419: {'lr': 4.136775747407145e-06, 'samples': 27152448, 'steps': 141418, 'loss/train': 1.568004846572876}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:27 - INFO - __main__ - Step 141423: {'lr': 4.132931069275786e-06, 'samples': 27153216, 'steps': 141422, 'loss/train': 1.0972918272018433}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:29 - INFO - __main__ - Step 141428: {'lr': 4.128127714252777e-06, 'samples': 27154176, 'steps': 141427, 'loss/train': 1.4460787773132324}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:31 - INFO - __main__ - Step 141432: {'lr': 4.124287024381057e-06, 'samples': 27154944, 'steps': 141431, 'loss/train': 1.165781021118164}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:33 - INFO - __main__ - Step 141436: {'lr': 4.120448107114177e-06, 'samples': 27155712, 'steps': 141435, 'loss/train': 1.3125170469284058}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:33 - INFO - __main__ - Step 141436: {'lr': 4.120448107114177e-06, 'samples': 27155712, 'steps': 141435, 'loss/train': 1.3125170469284058}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:37 - INFO - __main__ - Step 141443: {'lr': 4.1137342673107366e-06, 'samples': 27157056, 'steps': 141442, 'loss/train': 0.9833778142929077}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:39 - INFO - __main__ - Step 141448: {'lr': 4.108941991219222e-06, 'samples': 27158016, 'steps': 141447, 'loss/train': 0.9133830666542053}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:39 - INFO - __main__ - Step 141448: {'lr': 4.108941991219222e-06, 'samples': 27158016, 'steps': 141447, 'loss/train': 0.9133830666542053}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:43 - INFO - __main__ - Step 141455: {'lr': 4.102237458081249e-06, 'samples': 27159360, 'steps': 141454, 'loss/train': 2.7347512245178223}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:45 - INFO - __main__ - Step 141459: {'lr': 4.098408733828856e-06, 'samples': 27160128, 'steps': 141458, 'loss/train': 0.6587188839912415}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:47 - INFO - __main__ - Step 141464: {'lr': 4.093625321504507e-06, 'samples': 27161088, 'steps': 141463, 'loss/train': 1.0273503065109253}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:47 - INFO - __main__ - Step 141464: {'lr': 4.093625321504507e-06, 'samples': 27161088, 'steps': 141463, 'loss/train': 1.0273503065109253}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:51 - INFO - __main__ - Step 141472: {'lr': 4.085977623492082e-06, 'samples': 27162624, 'steps': 141471, 'loss/train': 1.3386622667312622}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:53 - INFO - __main__ - Step 141476: {'lr': 4.082156433793588e-06, 'samples': 27163392, 'steps': 141475, 'loss/train': 0.6175978779792786}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:55 - INFO - __main__ - Step 141480: {'lr': 4.078337017003692e-06, 'samples': 27164160, 'steps': 141479, 'loss/train': 1.0302541255950928}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:57 - INFO - __main__ - Step 141485: {'lr': 4.073565239210958e-06, 'samples': 27165120, 'steps': 141484, 'loss/train': 1.4073867797851562}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:09:59 - INFO - __main__ - Step 141489: {'lr': 4.069749811565965e-06, 'samples': 27165888, 'steps': 141488, 'loss/train': 1.430432915687561}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:02 - INFO - __main__ - Step 141493: {'lr': 4.0659361569190254e-06, 'samples': 27166656, 'steps': 141492, 'loss/train': 1.3772085905075073}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:04 - INFO - __main__ - Step 141497: {'lr': 4.062124275297563e-06, 'samples': 27167424, 'steps': 141496, 'loss/train': 1.3337675333023071}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:05 - INFO - __main__ - Step 141501: {'lr': 4.05831416672911e-06, 'samples': 27168192, 'steps': 141500, 'loss/train': 1.3665379285812378}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:07 - INFO - __main__ - Step 141505: {'lr': 4.05450583124109e-06, 'samples': 27168960, 'steps': 141504, 'loss/train': 1.278484582901001}}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:09 - INFO - __main__ - Step 141510: {'lr': 4.049747905317314e-06, 'samples': 27169920, 'steps': 141509, 'loss/train': 0.8817458748817444}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:12 - INFO - __main__ - Step 141515: {'lr': 4.044992749928228e-06, 'samples': 27170880, 'steps': 141514, 'loss/train': 1.5848987102508545}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:12 - INFO - __main__ - Step 141515: {'lr': 4.044992749928228e-06, 'samples': 27170880, 'steps': 141514, 'loss/train': 1.5848987102508545}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:15 - INFO - __main__ - Step 141522: {'lr': 4.038340186983791e-06, 'samples': 27172224, 'steps': 141521, 'loss/train': 1.4085183143615723}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:17 - INFO - __main__ - Step 141526: {'lr': 4.034541160618327e-06, 'samples': 27172992, 'steps': 141525, 'loss/train': 1.3692067861557007}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:17 - INFO - __main__ - Step 141526: {'lr': 4.034541160618327e-06, 'samples': 27172992, 'steps': 141525, 'loss/train': 1.3692067861557007}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:22 - INFO - __main__ - Step 141534: {'lr': 4.02694842767029e-06, 'samples': 27174528, 'steps': 141533, 'loss/train': 1.3013055324554443}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:22 - INFO - __main__ - Step 141534: {'lr': 4.02694842767029e-06, 'samples': 27174528, 'steps': 141533, 'loss/train': 1.3013055324554443}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:25 - INFO - __main__ - Step 141541: {'lr': 4.020310604995498e-06, 'samples': 27175872, 'steps': 141540, 'loss/train': 0.7951191663742065}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:27 - INFO - __main__ - Step 141546: {'lr': 4.015572628115943e-06, 'samples': 27176832, 'steps': 141545, 'loss/train': 0.9824565649032593}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:27 - INFO - __main__ - Step 141546: {'lr': 4.015572628115943e-06, 'samples': 27176832, 'steps': 141545, 'loss/train': 0.9824565649032593}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:32 - INFO - __main__ - Step 141554: {'lr': 4.007997628643817e-06, 'samples': 27178368, 'steps': 141553, 'loss/train': 1.100628137588501}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:33 - INFO - __main__ - Step 141558: {'lr': 4.004212789058909e-06, 'samples': 27179136, 'steps': 141557, 'loss/train': 1.0762543678283691}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:35 - INFO - __main__ - Step 141562: {'lr': 4.000429722944482e-06, 'samples': 27179904, 'steps': 141561, 'loss/train': 1.247206449508667}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:38 - INFO - __main__ - Step 141567: {'lr': 3.99570338428587e-06, 'samples': 27180864, 'steps': 141566, 'loss/train': 1.6075363159179688}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:40 - INFO - __main__ - Step 141571: {'lr': 3.991924308579753e-06, 'samples': 27181632, 'steps': 141570, 'loss/train': 1.4964455366134644}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:42 - INFO - __main__ - Step 141575: {'lr': 3.988147006432713e-06, 'samples': 27182400, 'steps': 141574, 'loss/train': 1.2750515937805176}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:43 - INFO - __main__ - Step 141579: {'lr': 3.984371477871979e-06, 'samples': 27183168, 'steps': 141578, 'loss/train': 1.5591577291488647}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:46 - INFO - __main__ - Step 141583: {'lr': 3.9805977229248054e-06, 'samples': 27183936, 'steps': 141582, 'loss/train': 0.6253988742828369}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:48 - INFO - __main__ - Step 141588: {'lr': 3.975883023426302e-06, 'samples': 27184896, 'steps': 141587, 'loss/train': 1.3857436180114746}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:48 - INFO - __main__ - Step 141588: {'lr': 3.975883023426302e-06, 'samples': 27184896, 'steps': 141587, 'loss/train': 1.3857436180114746}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:52 - INFO - __main__ - Step 141595: {'lr': 3.96928710003644e-06, 'samples': 27186240, 'steps': 141594, 'loss/train': 1.0960992574691772}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:54 - INFO - __main__ - Step 141599: {'lr': 3.965520439815323e-06, 'samples': 27187008, 'steps': 141598, 'loss/train': 1.3139533996582031}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:56 - INFO - __main__ - Step 141603: {'lr': 3.96175555334366e-06, 'samples': 27187776, 'steps': 141602, 'loss/train': 1.1527175903320312}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:58 - INFO - __main__ - Step 141608: {'lr': 3.957051939630557e-06, 'samples': 27188736, 'steps': 141607, 'loss/train': 1.3833152055740356}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:10:58 - INFO - __main__ - Step 141608: {'lr': 3.957051939630557e-06, 'samples': 27188736, 'steps': 141607, 'loss/train': 1.3833152055740356}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:01 - INFO - __main__ - Step 141614: {'lr': 3.951411261663623e-06, 'samples': 27189888, 'steps': 141613, 'loss/train': 0.5832651257514954}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:04 - INFO - __main__ - Step 141618: {'lr': 3.947653026993958e-06, 'samples': 27190656, 'steps': 141617, 'loss/train': 1.1039952039718628}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:04 - INFO - __main__ - Step 141618: {'lr': 3.947653026993958e-06, 'samples': 27190656, 'steps': 141617, 'loss/train': 1.1039952039718628}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:07 - INFO - __main__ - Step 141625: {'lr': 3.941080384732926e-06, 'samples': 27192000, 'steps': 141624, 'loss/train': 1.6318604946136475}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:10 - INFO - __main__ - Step 141630: {'lr': 3.936388966362836e-06, 'samples': 27192960, 'steps': 141629, 'loss/train': 1.0261729955673218}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:10 - INFO - __main__ - Step 141630: {'lr': 3.936388966362836e-06, 'samples': 27192960, 'steps': 141629, 'loss/train': 1.0261729955673218}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:14 - INFO - __main__ - Step 141638: {'lr': 3.928888462360919e-06, 'samples': 27194496, 'steps': 141637, 'loss/train': 1.4674416780471802}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:15 - INFO - __main__ - Step 141642: {'lr': 3.9251408713667505e-06, 'samples': 27195264, 'steps': 141641, 'loss/train': 1.3438849449157715}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:17 - INFO - __main__ - Step 141646: {'lr': 3.921395054413135e-06, 'samples': 27196032, 'steps': 141645, 'loss/train': 0.9496861100196838}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:20 - INFO - __main__ - Step 141651: {'lr': 3.916715278006744e-06, 'samples': 27196992, 'steps': 141650, 'loss/train': 1.0703130960464478}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:20 - INFO - __main__ - Step 141651: {'lr': 3.916715278006744e-06, 'samples': 27196992, 'steps': 141650, 'loss/train': 1.0703130960464478}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:24 - INFO - __main__ - Step 141659: {'lr': 3.909233401607648e-06, 'samples': 27198528, 'steps': 141658, 'loss/train': 1.1625169515609741}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:25 - INFO - __main__ - Step 141663: {'lr': 3.9054951246274835e-06, 'samples': 27199296, 'steps': 141662, 'loss/train': 1.2341045141220093}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:27 - INFO - __main__ - Step 141667: {'lr': 3.901758621829482e-06, 'samples': 27200064, 'steps': 141666, 'loss/train': 0.9676464200019836}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:30 - INFO - __main__ - Step 141672: {'lr': 3.897090488316712e-06, 'samples': 27201024, 'steps': 141671, 'loss/train': 1.465740442276001}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:32 - INFO - __main__ - Step 141676: {'lr': 3.893357977527101e-06, 'samples': 27201792, 'steps': 141675, 'loss/train': 1.0888111591339111}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:32 - INFO - __main__ - Step 141676: {'lr': 3.893357977527101e-06, 'samples': 27201792, 'steps': 141675, 'loss/train': 1.0888111591339111}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:35 - INFO - __main__ - Step 141683: {'lr': 3.886830352997861e-06, 'samples': 27203136, 'steps': 141682, 'loss/train': 1.0815876722335815}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:38 - INFO - __main__ - Step 141688: {'lr': 3.8821710908839295e-06, 'samples': 27204096, 'steps': 141687, 'loss/train': 1.80308997631073}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:40 - INFO - __main__ - Step 141693: {'lr': 3.877514601192345e-06, 'samples': 27205056, 'steps': 141692, 'loss/train': 1.4671849012374878}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:40 - INFO - __main__ - Step 141693: {'lr': 3.877514601192345e-06, 'samples': 27205056, 'steps': 141692, 'loss/train': 1.4671849012374878}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:43 - INFO - __main__ - Step 141700: {'lr': 3.871000173393674e-06, 'samples': 27206400, 'steps': 141699, 'loss/train': 1.3028454780578613}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:45 - INFO - __main__ - Step 141704: {'lr': 3.8672800830562016e-06, 'samples': 27207168, 'steps': 141703, 'loss/train': 1.114676594734192}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:45 - INFO - __main__ - Step 141704: {'lr': 3.8672800830562016e-06, 'samples': 27207168, 'steps': 141703, 'loss/train': 1.114676594734192}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:50 - INFO - __main__ - Step 141712: {'lr': 3.859845225781117e-06, 'samples': 27208704, 'steps': 141711, 'loss/train': 1.322885513305664}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:52 - INFO - __main__ - Step 141716: {'lr': 3.856130458897156e-06, 'samples': 27209472, 'steps': 141715, 'loss/train': 0.8814691305160522}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:54 - INFO - __main__ - Step 141720: {'lr': 3.852417466551211e-06, 'samples': 27210240, 'steps': 141719, 'loss/train': 0.5812214612960815}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:56 - INFO - __main__ - Step 141725: {'lr': 3.847778721603745e-06, 'samples': 27211200, 'steps': 141724, 'loss/train': 1.5728529691696167}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:11:58 - INFO - __main__ - Step 141729: {'lr': 3.844069722066329e-06, 'samples': 27211968, 'steps': 141728, 'loss/train': 1.0151349306106567}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:00 - INFO - __main__ - Step 141733: {'lr': 3.840362497153943e-06, 'samples': 27212736, 'steps': 141732, 'loss/train': 1.523417592048645}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:00 - INFO - __main__ - Step 141733: {'lr': 3.840362497153943e-06, 'samples': 27212736, 'steps': 141732, 'loss/train': 1.523417592048645}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:04 - INFO - __main__ - Step 141740: {'lr': 3.83387912382907e-06, 'samples': 27214080, 'steps': 141739, 'loss/train': 0.8294079303741455}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:06 - INFO - __main__ - Step 141745: {'lr': 3.829251470434059e-06, 'samples': 27215040, 'steps': 141744, 'loss/train': 0.9034472703933716}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:06 - INFO - __main__ - Step 141745: {'lr': 3.829251470434059e-06, 'samples': 27215040, 'steps': 141744, 'loss/train': 0.9034472703933716}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:06 - INFO - __main__ - Step 141745: {'lr': 3.829251470434059e-06, 'samples': 27215040, 'steps': 141744, 'loss/train': 0.9034472703933716}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:12 - INFO - __main__ - Step 141756: {'lr': 3.8190803940635624e-06, 'samples': 27217152, 'steps': 141755, 'loss/train': 0.8803466558456421}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:14 - INFO - __main__ - Step 141761: {'lr': 3.8144616145111276e-06, 'samples': 27218112, 'steps': 141760, 'loss/train': 0.8972194194793701}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:16 - INFO - __main__ - Step 141766: {'lr': 3.8098456081437917e-06, 'samples': 27219072, 'steps': 141765, 'loss/train': 1.2118955850601196}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:18 - INFO - __main__ - Step 141770: {'lr': 3.8061547997781943e-06, 'samples': 27219840, 'steps': 141769, 'loss/train': 1.5946654081344604}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:18 - INFO - __main__ - Step 141770: {'lr': 3.8061547997781943e-06, 'samples': 27219840, 'steps': 141769, 'loss/train': 1.5946654081344604}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:22 - INFO - __main__ - Step 141777: {'lr': 3.799700156003505e-06, 'samples': 27221184, 'steps': 141776, 'loss/train': 1.491884469985962}4}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:24 - INFO - __main__ - Step 141782: {'lr': 3.7950930241778536e-06, 'samples': 27222144, 'steps': 141781, 'loss/train': 1.3125410079956055}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:26 - INFO - __main__ - Step 141786: {'lr': 3.791409315565181e-06, 'samples': 27222912, 'steps': 141785, 'loss/train': 1.077893614768982}5}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:26 - INFO - __main__ - Step 141786: {'lr': 3.791409315565181e-06, 'samples': 27222912, 'steps': 141785, 'loss/train': 1.077893614768982}5}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:30 - INFO - __main__ - Step 141794: {'lr': 3.78404722338041e-06, 'samples': 27224448, 'steps': 141793, 'loss/train': 1.148646593093872}}5}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:32 - INFO - __main__ - Step 141798: {'lr': 3.7803688398613812e-06, 'samples': 27225216, 'steps': 141797, 'loss/train': 1.7014334201812744}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:34 - INFO - __main__ - Step 141802: {'lr': 3.7766922314265985e-06, 'samples': 27225984, 'steps': 141801, 'loss/train': 1.1601243019104004}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:36 - INFO - __main__ - Step 141807: {'lr': 3.772098967135762e-06, 'samples': 27226944, 'steps': 141806, 'loss/train': 1.2637841701507568}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:36 - INFO - __main__ - Step 141807: {'lr': 3.772098967135762e-06, 'samples': 27226944, 'steps': 141806, 'loss/train': 1.2637841701507568}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:40 - INFO - __main__ - Step 141815: {'lr': 3.7647555135095215e-06, 'samples': 27228480, 'steps': 141814, 'loss/train': 1.3283259868621826}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:42 - INFO - __main__ - Step 141819: {'lr': 3.7610864494784234e-06, 'samples': 27229248, 'steps': 141818, 'loss/train': 0.9595531821250916}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:44 - INFO - __main__ - Step 141823: {'lr': 3.757419160670572e-06, 'samples': 27230016, 'steps': 141822, 'loss/train': 1.5998033285140991}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:46 - INFO - __main__ - Step 141828: {'lr': 3.752837546108806e-06, 'samples': 27230976, 'steps': 141827, 'loss/train': 0.9706765413284302}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:46 - INFO - __main__ - Step 141828: {'lr': 3.752837546108806e-06, 'samples': 27230976, 'steps': 141827, 'loss/train': 0.9706765413284302}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:51 - INFO - __main__ - Step 141836: {'lr': 3.745512732500289e-06, 'samples': 27232512, 'steps': 141835, 'loss/train': 1.6456691026687622}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:53 - INFO - __main__ - Step 141840: {'lr': 3.7418529886861787e-06, 'samples': 27233280, 'steps': 141839, 'loss/train': 1.1947717666625977}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:54 - INFO - __main__ - Step 141844: {'lr': 3.7381950202340087e-06, 'samples': 27234048, 'steps': 141843, 'loss/train': 1.2517789602279663}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:54 - INFO - __main__ - Step 141844: {'lr': 3.7381950202340087e-06, 'samples': 27234048, 'steps': 141843, 'loss/train': 1.2517789602279663}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:12:54 - INFO - __main__ - Step 141844: {'lr': 3.7381950202340087e-06, 'samples': 27234048, 'steps': 141843, 'loss/train': 1.2517789602279663}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:00 - INFO - __main__ - Step 141854: {'lr': 3.7290578664850583e-06, 'samples': 27235968, 'steps': 141853, 'loss/train': 0.9180381894111633}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:03 - INFO - __main__ - Step 141860: {'lr': 3.7235809005718713e-06, 'samples': 27237120, 'steps': 141859, 'loss/train': 1.0947210788726807}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:05 - INFO - __main__ - Step 141864: {'lr': 3.7199318093247e-06, 'samples': 27237888, 'steps': 141863, 'loss/train': 1.4545807838439941}07}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:07 - INFO - __main__ - Step 141868: {'lr': 3.7162844935974825e-06, 'samples': 27238656, 'steps': 141867, 'loss/train': 0.8698520660400391}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:07 - INFO - __main__ - Step 141868: {'lr': 3.7162844935974825e-06, 'samples': 27238656, 'steps': 141867, 'loss/train': 0.8698520660400391}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:10 - INFO - __main__ - Step 141875: {'lr': 3.7099059634989053e-06, 'samples': 27240000, 'steps': 141874, 'loss/train': 0.8191941976547241}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:13 - INFO - __main__ - Step 141880: {'lr': 3.705353199798589e-06, 'samples': 27240960, 'steps': 141879, 'loss/train': 1.1412848234176636}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:15 - INFO - __main__ - Step 141884: {'lr': 3.701712986414085e-06, 'samples': 27241728, 'steps': 141883, 'loss/train': 1.0946468114852905}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:16 - INFO - __main__ - Step 141888: {'lr': 3.6980745486809296e-06, 'samples': 27242496, 'steps': 141887, 'loss/train': 1.4461404085159302}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:18 - INFO - __main__ - Step 141892: {'lr': 3.694437886625296e-06, 'samples': 27243264, 'steps': 141891, 'loss/train': 1.8673992156982422}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:21 - INFO - __main__ - Step 141897: {'lr': 3.6898945561422424e-06, 'samples': 27244224, 'steps': 141896, 'loss/train': 1.1957231760025024}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:23 - INFO - __main__ - Step 141901: {'lr': 3.6862618894569488e-06, 'samples': 27244992, 'steps': 141900, 'loss/train': 0.4576829969882965}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:25 - INFO - __main__ - Step 141905: {'lr': 3.6826309985343585e-06, 'samples': 27245760, 'steps': 141904, 'loss/train': 1.3304240703582764}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:27 - INFO - __main__ - Step 141909: {'lr': 3.6790018834006457e-06, 'samples': 27246528, 'steps': 141908, 'loss/train': 1.4143842458724976}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:28 - INFO - __main__ - Step 141913: {'lr': 3.675374544081983e-06, 'samples': 27247296, 'steps': 141912, 'loss/train': 1.310847520828247}6}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:31 - INFO - __main__ - Step 141918: {'lr': 3.6708428672134475e-06, 'samples': 27248256, 'steps': 141917, 'loss/train': 1.406219482421875}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:33 - INFO - __main__ - Step 141922: {'lr': 3.6672195235742667e-06, 'samples': 27249024, 'steps': 141921, 'loss/train': 1.5104260444641113}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:33 - INFO - __main__ - Step 141922: {'lr': 3.6672195235742667e-06, 'samples': 27249024, 'steps': 141921, 'loss/train': 1.5104260444641113}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:36 - INFO - __main__ - Step 141928: {'lr': 3.661787837936137e-06, 'samples': 27250176, 'steps': 141927, 'loss/train': 1.0329991579055786}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:38 - INFO - __main__ - Step 141933: {'lr': 3.6572644856293636e-06, 'samples': 27251136, 'steps': 141932, 'loss/train': 1.2877277135849}6}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:41 - INFO - __main__ - Step 141938: {'lr': 3.6527439082784665e-06, 'samples': 27252096, 'steps': 141937, 'loss/train': 1.0167546272277832}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:41 - INFO - __main__ - Step 141938: {'lr': 3.6527439082784665e-06, 'samples': 27252096, 'steps': 141937, 'loss/train': 1.0167546272277832}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:44 - INFO - __main__ - Step 141945: {'lr': 3.6464197620100783e-06, 'samples': 27253440, 'steps': 141944, 'loss/train': 1.1527694463729858}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:46 - INFO - __main__ - Step 141949: {'lr': 3.642808406202125e-06, 'samples': 27254208, 'steps': 141948, 'loss/train': 1.2649039030075073}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:48 - INFO - __main__ - Step 141954: {'lr': 3.638296709052108e-06, 'samples': 27255168, 'steps': 141953, 'loss/train': 1.5018678903579712}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:51 - INFO - __main__ - Step 141958: {'lr': 3.634689349451742e-06, 'samples': 27255936, 'steps': 141957, 'loss/train': 1.3791881799697876}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:53 - INFO - __main__ - Step 141962: {'lr': 3.631083765985865e-06, 'samples': 27256704, 'steps': 141961, 'loss/train': 1.5596235990524292}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:55 - INFO - __main__ - Step 141966: {'lr': 3.627479958680513e-06, 'samples': 27257472, 'steps': 141965, 'loss/train': 1.373599886894226}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:57 - INFO - __main__ - Step 141971: {'lr': 3.6229776973141228e-06, 'samples': 27258432, 'steps': 141970, 'loss/train': 1.1458240747451782}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:13:59 - INFO - __main__ - Step 141975: {'lr': 3.6193778864648805e-06, 'samples': 27259200, 'steps': 141974, 'loss/train': 0.3597372770309448}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:01 - INFO - __main__ - Step 141979: {'lr': 3.61577985186054e-06, 'samples': 27259968, 'steps': 141978, 'loss/train': 1.2715033292770386}8}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:03 - INFO - __main__ - Step 141983: {'lr': 3.61218359352708e-06, 'samples': 27260736, 'steps': 141982, 'loss/train': 1.6053742170333862}8}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:03 - INFO - __main__ - Step 141983: {'lr': 3.61218359352708e-06, 'samples': 27260736, 'steps': 141982, 'loss/train': 1.6053742170333862}8}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:06 - INFO - __main__ - Step 141990: {'lr': 3.605894415673133e-06, 'samples': 27262080, 'steps': 141989, 'loss/train': 1.3720098733901978}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:08 - INFO - __main__ - Step 141995: {'lr': 3.601405476410857e-06, 'samples': 27263040, 'steps': 141994, 'loss/train': 1.254608154296875}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:11 - INFO - __main__ - Step 142000: {'lr': 3.5969193127336762e-06, 'samples': 27264000, 'steps': 141999, 'loss/train': 0.7151772975921631}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:11 - INFO - __main__ - Step 142000: {'lr': 3.5969193127336762e-06, 'samples': 27264000, 'steps': 141999, 'loss/train': 0.7151772975921631}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:15 - INFO - __main__ - Step 142007: {'lr': 3.590643346664785e-06, 'samples': 27265344, 'steps': 142006, 'loss/train': 0.6447149515151978}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:16 - INFO - __main__ - Step 142011: {'lr': 3.587059522952435e-06, 'samples': 27266112, 'steps': 142010, 'loss/train': 1.598928451538086}}}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:18 - INFO - __main__ - Step 142015: {'lr': 3.5834774757179665e-06, 'samples': 27266880, 'steps': 142014, 'loss/train': 1.3470423221588135}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:21 - INFO - __main__ - Step 142020: {'lr': 3.5790024148862345e-06, 'samples': 27267840, 'steps': 142019, 'loss/train': 1.5563069581985474}}tils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:23 - INFO - __main__ - Step 142024: {'lr': 3.5754243648214445e-06, 'samples': 27268608, 'steps': 142023, 'loss/train': 0.050808753818273544}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:25 - INFO - __main__ - Step 142028: {'lr': 3.5718480913184136e-06, 'samples': 27269376, 'steps': 142027, 'loss/train': 1.416524052619934}44}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:26 - INFO - __main__ - Step 142032: {'lr': 3.5682735944029542e-06, 'samples': 27270144, 'steps': 142031, 'loss/train': 1.0735936164855957}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:29 - INFO - __main__ - Step 142036: {'lr': 3.5647008741008235e-06, 'samples': 27270912, 'steps': 142035, 'loss/train': 1.8481760025024414}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:32 - INFO - __main__ - Step 142041: {'lr': 3.560237472124889e-06, 'samples': 27271872, 'steps': 142040, 'loss/train': 1.2404077053070068}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:34 - INFO - __main__ - Step 142046: {'lr': 3.555776846197817e-06, 'samples': 27272832, 'steps': 142045, 'loss/train': 1.7292296886444092}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:36 - INFO - __main__ - Step 142050: {'lr': 3.552210344245105e-06, 'samples': 27273600, 'steps': 142049, 'loss/train': 0.9861950278282166}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:39 - INFO - __main__ - Step 142055: {'lr': 3.5477547153325573e-06, 'samples': 27274560, 'steps': 142054, 'loss/train': 1.2517012357711792}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:39 - INFO - __main__ - Step 142055: {'lr': 3.5477547153325573e-06, 'samples': 27274560, 'steps': 142054, 'loss/train': 1.2517012357711792}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:42 - INFO - __main__ - Step 142061: {'lr': 3.5424116252119885e-06, 'samples': 27275712, 'steps': 142060, 'loss/train': 1.2985634803771973}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:45 - INFO - __main__ - Step 142066: {'lr': 3.537962103982495e-06, 'samples': 27276672, 'steps': 142065, 'loss/train': 1.197359323501587}3}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:47 - INFO - __main__ - Step 142071: {'lr': 3.533515359052747e-06, 'samples': 27277632, 'steps': 142070, 'loss/train': 1.1298803091049194}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:47 - INFO - __main__ - Step 142071: {'lr': 3.533515359052747e-06, 'samples': 27277632, 'steps': 142070, 'loss/train': 1.1298803091049194}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:50 - INFO - __main__ - Step 142078: {'lr': 3.5272945804301347e-06, 'samples': 27278976, 'steps': 142077, 'loss/train': 1.3799455165863037}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:53 - INFO - __main__ - Step 142082: {'lr': 3.5237422930293474e-06, 'samples': 27279744, 'steps': 142081, 'loss/train': 1.3829742670059204}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:55 - INFO - __main__ - Step 142087: {'lr': 3.5193044325953772e-06, 'samples': 27280704, 'steps': 142086, 'loss/train': 1.2709687948226929}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:55 - INFO - __main__ - Step 142087: {'lr': 3.5193044325953772e-06, 'samples': 27280704, 'steps': 142086, 'loss/train': 1.2709687948226929}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:14:58 - INFO - __main__ - Step 142093: {'lr': 3.5139826650721073e-06, 'samples': 27281856, 'steps': 142092, 'loss/train': 1.1426702737808228}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:01 - INFO - __main__ - Step 142098: {'lr': 3.5095509130259327e-06, 'samples': 27282816, 'steps': 142097, 'loss/train': 1.0334078073501587}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:03 - INFO - __main__ - Step 142102: {'lr': 3.5060075105528833e-06, 'samples': 27283584, 'steps': 142101, 'loss/train': 1.4743255376815796}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:05 - INFO - __main__ - Step 142106: {'lr': 3.502465885142053e-06, 'samples': 27284352, 'steps': 142105, 'loss/train': 1.4503601789474487}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:06 - INFO - __main__ - Step 142110: {'lr': 3.498926036818978e-06, 'samples': 27285120, 'steps': 142109, 'loss/train': 1.4366575479507446}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:08 - INFO - __main__ - Step 142114: {'lr': 3.4953879656091923e-06, 'samples': 27285888, 'steps': 142113, 'loss/train': 1.372241497039795}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:11 - INFO - __main__ - Step 142119: {'lr': 3.490967875701384e-06, 'samples': 27286848, 'steps': 142118, 'loss/train': 1.0834412574768066}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:11 - INFO - __main__ - Step 142119: {'lr': 3.490967875701384e-06, 'samples': 27286848, 'steps': 142118, 'loss/train': 1.0834412574768066}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:15 - INFO - __main__ - Step 142127: {'lr': 3.4839015076741644e-06, 'samples': 27288384, 'steps': 142126, 'loss/train': 1.5727320909500122}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:15 - INFO - __main__ - Step 142127: {'lr': 3.4839015076741644e-06, 'samples': 27288384, 'steps': 142126, 'loss/train': 1.5727320909500122}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:18 - INFO - __main__ - Step 142134: {'lr': 3.4777242671517885e-06, 'samples': 27289728, 'steps': 142133, 'loss/train': 1.3558086156845093}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:21 - INFO - __main__ - Step 142139: {'lr': 3.4733152848589742e-06, 'samples': 27290688, 'steps': 142138, 'loss/train': 1.084580659866333}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:23 - INFO - __main__ - Step 142144: {'lr': 3.4689090795941847e-06, 'samples': 27291648, 'steps': 142143, 'loss/train': 1.0845856666564941}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:23 - INFO - __main__ - Step 142144: {'lr': 3.4689090795941847e-06, 'samples': 27291648, 'steps': 142143, 'loss/train': 1.0845856666564941}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:27 - INFO - __main__ - Step 142151: {'lr': 3.4627450577249808e-06, 'samples': 27292992, 'steps': 142150, 'loss/train': 1.762975811958313}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:29 - INFO - __main__ - Step 142155: {'lr': 3.4592252033947003e-06, 'samples': 27293760, 'steps': 142154, 'loss/train': 1.398964762687683}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:31 - INFO - __main__ - Step 142160: {'lr': 3.4548278849528823e-06, 'samples': 27294720, 'steps': 142159, 'loss/train': 1.0042423009872437}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:33 - INFO - __main__ - Step 142164: {'lr': 3.451312029807141e-06, 'samples': 27295488, 'steps': 142163, 'loss/train': 0.9495099186897278}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:35 - INFO - __main__ - Step 142168: {'lr': 3.4477979521179426e-06, 'samples': 27296256, 'steps': 142167, 'loss/train': 1.4838839769363403}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:37 - INFO - __main__ - Step 142172: {'lr': 3.4442856519106556e-06, 'samples': 27297024, 'steps': 142171, 'loss/train': 1.7813079357147217}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:39 - INFO - __main__ - Step 142176: {'lr': 3.440775129210538e-06, 'samples': 27297792, 'steps': 142175, 'loss/train': 1.1221270561218262}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:41 - INFO - __main__ - Step 142180: {'lr': 3.4372663840429587e-06, 'samples': 27298560, 'steps': 142179, 'loss/train': 1.0081218481063843}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:44 - INFO - __main__ - Step 142185: {'lr': 3.4328829522771168e-06, 'samples': 27299520, 'steps': 142184, 'loss/train': 1.491267442703247}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:44 - INFO - __main__ - Step 142185: {'lr': 3.4328829522771168e-06, 'samples': 27299520, 'steps': 142184, 'loss/train': 1.491267442703247}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:47 - INFO - __main__ - Step 142192: {'lr': 3.426750813988161e-06, 'samples': 27300864, 'steps': 142191, 'loss/train': 0.1290620118379593}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:49 - INFO - __main__ - Step 142196: {'lr': 3.423249179203447e-06, 'samples': 27301632, 'steps': 142195, 'loss/train': 0.8834850788116455}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:51 - INFO - __main__ - Step 142201: {'lr': 3.4188746355582887e-06, 'samples': 27302592, 'steps': 142200, 'loss/train': 1.1848098039627075}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:54 - INFO - __main__ - Step 142205: {'lr': 3.4153770005415085e-06, 'samples': 27303360, 'steps': 142204, 'loss/train': 1.1984400749206543}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:56 - INFO - __main__ - Step 142209: {'lr': 3.411881143240314e-06, 'samples': 27304128, 'steps': 142208, 'loss/train': 1.6009880304336548}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:57 - INFO - __main__ - Step 142213: {'lr': 3.408387063679991e-06, 'samples': 27304896, 'steps': 142212, 'loss/train': 1.2755959033966064}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:59 - INFO - __main__ - Step 142217: {'lr': 3.4048947618856294e-06, 'samples': 27305664, 'steps': 142216, 'loss/train': 1.4316691160202026}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:15:59 - INFO - __main__ - Step 142217: {'lr': 3.4048947618856294e-06, 'samples': 27305664, 'steps': 142216, 'loss/train': 1.4316691160202026}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:04 - INFO - __main__ - Step 142225: {'lr': 3.3979154916956856e-06, 'samples': 27307200, 'steps': 142224, 'loss/train': 1.1542023420333862}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:06 - INFO - __main__ - Step 142229: {'lr': 3.394428523350368e-06, 'samples': 27307968, 'steps': 142228, 'loss/train': 0.9675542712211609}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:07 - INFO - __main__ - Step 142233: {'lr': 3.39094333287171e-06, 'samples': 27308736, 'steps': 142232, 'loss/train': 1.2987756729125977}}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:09 - INFO - __main__ - Step 142238: {'lr': 3.38658934493663e-06, 'samples': 27309696, 'steps': 142237, 'loss/train': 1.639046549797058}}}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:12 - INFO - __main__ - Step 142242: {'lr': 3.383108154749737e-06, 'samples': 27310464, 'steps': 142241, 'loss/train': 1.3971501588821411}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:14 - INFO - __main__ - Step 142246: {'lr': 3.3796287425111315e-06, 'samples': 27311232, 'steps': 142245, 'loss/train': 1.6936745643615723}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:14 - INFO - __main__ - Step 142246: {'lr': 3.3796287425111315e-06, 'samples': 27311232, 'steps': 142245, 'loss/train': 1.6936745643615723}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:17 - INFO - __main__ - Step 142253: {'lr': 3.373544049356997e-06, 'samples': 27312576, 'steps': 142252, 'loss/train': 1.2161318063735962}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:19 - INFO - __main__ - Step 142258: {'lr': 3.36920117373557e-06, 'samples': 27313536, 'steps': 142257, 'loss/train': 1.3103506565093994}}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:22 - INFO - __main__ - Step 142263: {'lr': 3.364861076314979e-06, 'samples': 27314496, 'steps': 142262, 'loss/train': 0.9303818345069885}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:22 - INFO - __main__ - Step 142263: {'lr': 3.364861076314979e-06, 'samples': 27314496, 'steps': 142262, 'loss/train': 0.9303818345069885}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:25 - INFO - __main__ - Step 142270: {'lr': 3.3587896073965783e-06, 'samples': 27315840, 'steps': 142269, 'loss/train': 1.061408519744873}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:27 - INFO - __main__ - Step 142274: {'lr': 3.355322641497377e-06, 'samples': 27316608, 'steps': 142273, 'loss/train': 1.0634902715682983}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:29 - INFO - __main__ - Step 142279: {'lr': 3.35099143464776e-06, 'samples': 27317568, 'steps': 142278, 'loss/train': 1.1415417194366455}}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:32 - INFO - __main__ - Step 142283: {'lr': 3.347528469617972e-06, 'samples': 27318336, 'steps': 142282, 'loss/train': 0.982093334197998}}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:34 - INFO - __main__ - Step 142287: {'lr': 3.3440672827929342e-06, 'samples': 27319104, 'steps': 142286, 'loss/train': 1.386486291885376}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:35 - INFO - __main__ - Step 142291: {'lr': 3.3406078741976266e-06, 'samples': 27319872, 'steps': 142290, 'loss/train': 1.233959674835205}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:37 - INFO - __main__ - Step 142295: {'lr': 3.3371502438569733e-06, 'samples': 27320640, 'steps': 142294, 'loss/train': 1.165158748626709}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:40 - INFO - __main__ - Step 142300: {'lr': 3.332830706639728e-06, 'samples': 27321600, 'steps': 142299, 'loss/train': 1.2554991245269775}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:40 - INFO - __main__ - Step 142300: {'lr': 3.332830706639728e-06, 'samples': 27321600, 'steps': 142299, 'loss/train': 1.2554991245269775}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:44 - INFO - __main__ - Step 142307: {'lr': 3.3267880226121317e-06, 'samples': 27322944, 'steps': 142306, 'loss/train': 1.1694025993347168}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:45 - INFO - __main__ - Step 142311: {'lr': 3.3233375055392313e-06, 'samples': 27323712, 'steps': 142310, 'loss/train': 1.1741605997085571}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:47 - INFO - __main__ - Step 142315: {'lr': 3.319888766845469e-06, 'samples': 27324480, 'steps': 142314, 'loss/train': 1.395835041999817}1}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:50 - INFO - __main__ - Step 142319: {'lr': 3.31644180655577e-06, 'samples': 27325248, 'steps': 142318, 'loss/train': 1.319704532623291}}1}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:52 - INFO - __main__ - Step 142323: {'lr': 3.3129966246949193e-06, 'samples': 27326016, 'steps': 142322, 'loss/train': 1.2959647178649902}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:53 - INFO - __main__ - Step 142327: {'lr': 3.3095532212877867e-06, 'samples': 27326784, 'steps': 142326, 'loss/train': 1.3063195943832397}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:53 - INFO - __main__ - Step 142327: {'lr': 3.3095532212877867e-06, 'samples': 27326784, 'steps': 142326, 'loss/train': 1.3063195943832397}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:57 - INFO - __main__ - Step 142334: {'lr': 3.303531544804256e-06, 'samples': 27328128, 'steps': 142333, 'loss/train': 1.6379519701004028}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:16:59 - INFO - __main__ - Step 142338: {'lr': 3.3000930322727998e-06, 'samples': 27328896, 'steps': 142337, 'loss/train': 1.5356707572937012}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:02 - INFO - __main__ - Step 142343: {'lr': 3.2957973926927287e-06, 'samples': 27329856, 'steps': 142342, 'loss/train': 1.0783246755599976}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:04 - INFO - __main__ - Step 142347: {'lr': 3.292362881926303e-06, 'samples': 27330624, 'steps': 142346, 'loss/train': 1.2995859384536743}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:06 - INFO - __main__ - Step 142351: {'lr': 3.2889301497623093e-06, 'samples': 27331392, 'steps': 142350, 'loss/train': 1.7253361940383911}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:08 - INFO - __main__ - Step 142355: {'lr': 3.285499196225561e-06, 'samples': 27332160, 'steps': 142354, 'loss/train': 1.399717092514038}1}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:09 - INFO - __main__ - Step 142359: {'lr': 3.282070021340733e-06, 'samples': 27332928, 'steps': 142358, 'loss/train': 1.2036041021347046}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:11 - INFO - __main__ - Step 142363: {'lr': 3.278642625132583e-06, 'samples': 27333696, 'steps': 142362, 'loss/train': 1.4713197946548462}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:14 - INFO - __main__ - Step 142368: {'lr': 3.2743608811741375e-06, 'samples': 27334656, 'steps': 142367, 'loss/train': 1.1317704916000366}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:14 - INFO - __main__ - Step 142368: {'lr': 3.2743608811741375e-06, 'samples': 27334656, 'steps': 142367, 'loss/train': 1.1317704916000366}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:17 - INFO - __main__ - Step 142375: {'lr': 3.2683711088152825e-06, 'samples': 27336000, 'steps': 142374, 'loss/train': 1.2354097366333008}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:19 - INFO - __main__ - Step 142379: {'lr': 3.2649508275607863e-06, 'samples': 27336768, 'steps': 142378, 'loss/train': 1.3069185018539429}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:19 - INFO - __main__ - Step 142379: {'lr': 3.2649508275607863e-06, 'samples': 27336768, 'steps': 142378, 'loss/train': 1.3069185018539429}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:23 - INFO - __main__ - Step 142387: {'lr': 3.258115601476763e-06, 'samples': 27338304, 'steps': 142386, 'loss/train': 1.3409656286239624}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:23 - INFO - __main__ - Step 142387: {'lr': 3.258115601476763e-06, 'samples': 27338304, 'steps': 142386, 'loss/train': 1.3409656286239624}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:28 - INFO - __main__ - Step 142395: {'lr': 3.251287490790178e-06, 'samples': 27339840, 'steps': 142394, 'loss/train': 1.2516003847122192}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:30 - INFO - __main__ - Step 142399: {'lr': 3.2478761037824934e-06, 'samples': 27340608, 'steps': 142398, 'loss/train': 1.0817911624908447}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:32 - INFO - __main__ - Step 142403: {'lr': 3.244466495697984e-06, 'samples': 27341376, 'steps': 142402, 'loss/train': 1.2576797008514404}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:34 - INFO - __main__ - Step 142407: {'lr': 3.241058666561242e-06, 'samples': 27342144, 'steps': 142406, 'loss/train': 1.29060697555542}4}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:36 - INFO - __main__ - Step 142412: {'lr': 3.2368013818230567e-06, 'samples': 27343104, 'steps': 142411, 'loss/train': 1.1243236064910889}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:36 - INFO - __main__ - Step 142412: {'lr': 3.2368013818230567e-06, 'samples': 27343104, 'steps': 142411, 'loss/train': 1.1243236064910889}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:39 - INFO - __main__ - Step 142418: {'lr': 3.2316963093352027e-06, 'samples': 27344256, 'steps': 142417, 'loss/train': 0.8412118554115295}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:41 - INFO - __main__ - Step 142423: {'lr': 3.2274451399831873e-06, 'samples': 27345216, 'steps': 142422, 'loss/train': 1.5264198780059814}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:44 - INFO - __main__ - Step 142428: {'lr': 3.223196750428814e-06, 'samples': 27346176, 'steps': 142427, 'loss/train': 1.3391939401626587}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:44 - INFO - __main__ - Step 142428: {'lr': 3.223196750428814e-06, 'samples': 27346176, 'steps': 142427, 'loss/train': 1.3391939401626587}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:48 - INFO - __main__ - Step 142435: {'lr': 3.217253675203785e-06, 'samples': 27347520, 'steps': 142434, 'loss/train': 1.0343519449234009}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:49 - INFO - __main__ - Step 142439: {'lr': 3.21386007853261e-06, 'samples': 27348288, 'steps': 142438, 'loss/train': 1.2293931245803833}}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:51 - INFO - __main__ - Step 142443: {'lr': 3.2104682610298308e-06, 'samples': 27349056, 'steps': 142442, 'loss/train': 1.7400109767913818}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:54 - INFO - __main__ - Step 142448: {'lr': 3.2062309911442266e-06, 'samples': 27350016, 'steps': 142447, 'loss/train': 1.4622310400009155}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:54 - INFO - __main__ - Step 142448: {'lr': 3.2062309911442266e-06, 'samples': 27350016, 'steps': 142447, 'loss/train': 1.4622310400009155}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:58 - INFO - __main__ - Step 142455: {'lr': 3.2003034837765345e-06, 'samples': 27351360, 'steps': 142454, 'loss/train': 1.470068335533142}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:59 - INFO - __main__ - Step 142459: {'lr': 3.1969187831918656e-06, 'samples': 27352128, 'steps': 142458, 'loss/train': 1.147691249847412}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:17:59 - INFO - __main__ - Step 142459: {'lr': 3.1969187831918656e-06, 'samples': 27352128, 'steps': 142458, 'loss/train': 1.147691249847412}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:03 - INFO - __main__ - Step 142465: {'lr': 3.191845068492266e-06, 'samples': 27353280, 'steps': 142464, 'loss/train': 1.4207849502563477}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:03 - INFO - __main__ - Step 142465: {'lr': 3.191845068492266e-06, 'samples': 27353280, 'steps': 142464, 'loss/train': 1.4207849502563477}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:07 - INFO - __main__ - Step 142474: {'lr': 3.1842420030049335e-06, 'samples': 27355008, 'steps': 142473, 'loss/train': 1.4078844785690308}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:09 - INFO - __main__ - Step 142478: {'lr': 3.1808657542675146e-06, 'samples': 27355776, 'steps': 142477, 'loss/train': 0.8735725283622742}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:12 - INFO - __main__ - Step 142483: {'lr': 3.176647945638711e-06, 'samples': 27356736, 'steps': 142482, 'loss/train': 1.4344313144683838}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:14 - INFO - __main__ - Step 142487: {'lr': 3.173275700599637e-06, 'samples': 27357504, 'steps': 142486, 'loss/train': 1.6038118600845337}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:16 - INFO - __main__ - Step 142491: {'lr': 3.169905235021614e-06, 'samples': 27358272, 'steps': 142490, 'loss/train': 1.1405991315841675}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:17 - INFO - __main__ - Step 142495: {'lr': 3.166536548928872e-06, 'samples': 27359040, 'steps': 142494, 'loss/train': 1.4712145328521729}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:17 - INFO - __main__ - Step 142495: {'lr': 3.166536548928872e-06, 'samples': 27359040, 'steps': 142494, 'loss/train': 1.4712145328521729}}4}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:21 - INFO - __main__ - Step 142502: {'lr': 3.1606456302261667e-06, 'samples': 27360384, 'steps': 142501, 'loss/train': 0.06121412664651871}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:23 - INFO - __main__ - Step 142506: {'lr': 3.1572818378432233e-06, 'samples': 27361152, 'steps': 142505, 'loss/train': 1.5785415172576904}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:26 - INFO - __main__ - Step 142511: {'lr': 3.15307959989658e-06, 'samples': 27362112, 'steps': 142510, 'loss/train': 0.6908512115478516}4}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:28 - INFO - __main__ - Step 142515: {'lr': 3.1497198115944258e-06, 'samples': 27362880, 'steps': 142514, 'loss/train': 1.3756153583526611}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:28 - INFO - __main__ - Step 142515: {'lr': 3.1497198115944258e-06, 'samples': 27362880, 'steps': 142514, 'loss/train': 1.3756153583526611}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:31 - INFO - __main__ - Step 142522: {'lr': 3.1438444643169252e-06, 'samples': 27364224, 'steps': 142521, 'loss/train': 1.2653898000717163}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:33 - INFO - __main__ - Step 142526: {'lr': 3.1404895700578027e-06, 'samples': 27364992, 'steps': 142525, 'loss/train': 1.3266955614089966}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:36 - INFO - __main__ - Step 142532: {'lr': 3.1354605656092605e-06, 'samples': 27366144, 'steps': 142531, 'loss/train': 0.7775653600692749}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:38 - INFO - __main__ - Step 142536: {'lr': 3.1321101206389092e-06, 'samples': 27366912, 'steps': 142535, 'loss/train': 1.5170024633407593}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:38 - INFO - __main__ - Step 142536: {'lr': 3.1321101206389092e-06, 'samples': 27366912, 'steps': 142535, 'loss/train': 1.5170024633407593}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:42 - INFO - __main__ - Step 142543: {'lr': 3.1262511244974923e-06, 'samples': 27368256, 'steps': 142542, 'loss/train': 1.3106626272201538}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:44 - INFO - __main__ - Step 142547: {'lr': 3.122905573919144e-06, 'samples': 27369024, 'steps': 142546, 'loss/train': 1.6273276805877686}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:46 - INFO - __main__ - Step 142552: {'lr': 3.1187261385767494e-06, 'samples': 27369984, 'steps': 142551, 'loss/train': 1.4677269458770752}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:46 - INFO - __main__ - Step 142552: {'lr': 3.1187261385767494e-06, 'samples': 27369984, 'steps': 142551, 'loss/train': 1.4677269458770752}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:49 - INFO - __main__ - Step 142558: {'lr': 3.1137144871194177e-06, 'samples': 27371136, 'steps': 142557, 'loss/train': 1.0915354490280151}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:51 - INFO - __main__ - Step 142562: {'lr': 3.1103756110058835e-06, 'samples': 27371904, 'steps': 142561, 'loss/train': 1.067132830619812}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:54 - INFO - __main__ - Step 142567: {'lr': 3.1062045188715305e-06, 'samples': 27372864, 'steps': 142566, 'loss/train': 1.3246915340423584}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:56 - INFO - __main__ - Step 142571: {'lr': 3.102869647599371e-06, 'samples': 27373632, 'steps': 142570, 'loss/train': 1.7210301160812378}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:58 - INFO - __main__ - Step 142575: {'lr': 3.0995365562958565e-06, 'samples': 27374400, 'steps': 142574, 'loss/train': 1.3934166431427002}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:18:59 - INFO - __main__ - Step 142579: {'lr': 3.096205244984912e-06, 'samples': 27375168, 'steps': 142578, 'loss/train': 1.2635935544967651}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:01 - INFO - __main__ - Step 142583: {'lr': 3.0928757136906295e-06, 'samples': 27375936, 'steps': 142582, 'loss/train': 1.5805730819702148}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:04 - INFO - __main__ - Step 142588: {'lr': 3.0887163027577546e-06, 'samples': 27376896, 'steps': 142587, 'loss/train': 1.3346433639526367}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:04 - INFO - __main__ - Step 142588: {'lr': 3.0887163027577546e-06, 'samples': 27376896, 'steps': 142587, 'loss/train': 1.3346433639526367}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:07 - INFO - __main__ - Step 142594: {'lr': 3.0837286810380673e-06, 'samples': 27378048, 'steps': 142593, 'loss/train': 0.9343197345733643}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:07 - INFO - __main__ - Step 142594: {'lr': 3.0837286810380673e-06, 'samples': 27378048, 'steps': 142593, 'loss/train': 0.9343197345733643}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:12 - INFO - __main__ - Step 142603: {'lr': 3.0762547583085e-06, 'samples': 27379776, 'steps': 142602, 'loss/train': 1.4612998962402344}43}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:14 - INFO - __main__ - Step 142607: {'lr': 3.0729359076176464e-06, 'samples': 27380544, 'steps': 142606, 'loss/train': 1.039892554283142}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:16 - INFO - __main__ - Step 142611: {'lr': 3.0696188371112377e-06, 'samples': 27381312, 'steps': 142610, 'loss/train': 1.1267673969268799}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:18 - INFO - __main__ - Step 142615: {'lr': 3.066303546813115e-06, 'samples': 27382080, 'steps': 142614, 'loss/train': 1.3153598308563232}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:18 - INFO - __main__ - Step 142615: {'lr': 3.066303546813115e-06, 'samples': 27382080, 'steps': 142614, 'loss/train': 1.3153598308563232}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:18 - INFO - __main__ - Step 142615: {'lr': 3.066303546813115e-06, 'samples': 27382080, 'steps': 142614, 'loss/train': 1.3153598308563232}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:23 - INFO - __main__ - Step 142626: {'lr': 3.057195677887481e-06, 'samples': 27384192, 'steps': 142625, 'loss/train': 0.7727726697921753}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:26 - INFO - __main__ - Step 142631: {'lr': 3.0530601881815776e-06, 'samples': 27385152, 'steps': 142630, 'loss/train': 1.0430774688720703}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:28 - INFO - __main__ - Step 142636: {'lr': 3.04892748023769e-06, 'samples': 27386112, 'steps': 142635, 'loss/train': 1.311407208442688}03}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:30 - INFO - __main__ - Step 142640: {'lr': 3.0456233167824865e-06, 'samples': 27386880, 'steps': 142639, 'loss/train': 1.5400511026382446}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:30 - INFO - __main__ - Step 142640: {'lr': 3.0456233167824865e-06, 'samples': 27386880, 'steps': 142639, 'loss/train': 1.5400511026382446}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:34 - INFO - __main__ - Step 142647: {'lr': 3.0398453147927605e-06, 'samples': 27388224, 'steps': 142646, 'loss/train': 1.5008466243743896}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:36 - INFO - __main__ - Step 142652: {'lr': 3.0357215087999655e-06, 'samples': 27389184, 'steps': 142651, 'loss/train': 1.1874009370803833}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:36 - INFO - __main__ - Step 142652: {'lr': 3.0357215087999655e-06, 'samples': 27389184, 'steps': 142651, 'loss/train': 1.1874009370803833}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:40 - INFO - __main__ - Step 142660: {'lr': 3.029129205702058e-06, 'samples': 27390720, 'steps': 142659, 'loss/train': 0.99903404712677}33}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:40 - INFO - __main__ - Step 142660: {'lr': 3.029129205702058e-06, 'samples': 27390720, 'steps': 142659, 'loss/train': 0.99903404712677}33}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:40 - INFO - __main__ - Step 142660: {'lr': 3.029129205702058e-06, 'samples': 27390720, 'steps': 142659, 'loss/train': 0.99903404712677}33}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:46 - INFO - __main__ - Step 142670: {'lr': 3.0208988421656093e-06, 'samples': 27392640, 'steps': 142669, 'loss/train': 1.6402387619018555}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:48 - INFO - __main__ - Step 142674: {'lr': 3.0176098127094876e-06, 'samples': 27393408, 'steps': 142673, 'loss/train': 0.21918943524360657}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:50 - INFO - __main__ - Step 142679: {'lr': 3.0135010298373012e-06, 'samples': 27394368, 'steps': 142678, 'loss/train': 2.0655648708343506}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:53 - INFO - __main__ - Step 142683: {'lr': 3.0102160067268515e-06, 'samples': 27395136, 'steps': 142682, 'loss/train': 1.4866403341293335}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:55 - INFO - __main__ - Step 142687: {'lr': 3.006932764252929e-06, 'samples': 27395904, 'steps': 142686, 'loss/train': 1.525759220123291}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:56 - INFO - __main__ - Step 142691: {'lr': 3.0036513024392644e-06, 'samples': 27396672, 'steps': 142690, 'loss/train': 1.3295998573303223}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:19:58 - INFO - __main__ - Step 142695: {'lr': 3.0003716213095054e-06, 'samples': 27397440, 'steps': 142694, 'loss/train': 1.4845733642578125}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:01 - INFO - __main__ - Step 142699: {'lr': 2.9970937208872727e-06, 'samples': 27398208, 'steps': 142698, 'loss/train': 1.0926518440246582}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:02 - INFO - __main__ - Step 142703: {'lr': 2.9938176011961858e-06, 'samples': 27398976, 'steps': 142702, 'loss/train': 1.5019415616989136}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:04 - INFO - __main__ - Step 142707: {'lr': 2.9905432622598926e-06, 'samples': 27399744, 'steps': 142706, 'loss/train': 1.6209896802902222}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:07 - INFO - __main__ - Step 142712: {'lr': 2.986452842811932e-06, 'samples': 27400704, 'steps': 142711, 'loss/train': 0.976443350315094}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:09 - INFO - __main__ - Step 142716: {'lr': 2.9831825106602096e-06, 'samples': 27401472, 'steps': 142715, 'loss/train': 1.1813008785247803}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:09 - INFO - __main__ - Step 142716: {'lr': 2.9831825106602096e-06, 'samples': 27401472, 'steps': 142715, 'loss/train': 1.1813008785247803}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:12 - INFO - __main__ - Step 142724: {'lr': 2.9766471888747204e-06, 'samples': 27403008, 'steps': 142723, 'loss/train': 1.3307499885559082}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:14 - INFO - __main__ - Step 142728: {'lr': 2.9733821992880273e-06, 'samples': 27403776, 'steps': 142727, 'loss/train': 1.534027099609375}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:14 - INFO - __main__ - Step 142728: {'lr': 2.9733821992880273e-06, 'samples': 27403776, 'steps': 142727, 'loss/train': 1.534027099609375}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:18 - INFO - __main__ - Step 142736: {'lr': 2.966857562844566e-06, 'samples': 27405312, 'steps': 142735, 'loss/train': 1.4415359497070312}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:20 - INFO - __main__ - Step 142740: {'lr': 2.963597916034816e-06, 'samples': 27406080, 'steps': 142739, 'loss/train': 1.3308643102645874}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:22 - INFO - __main__ - Step 142744: {'lr': 2.9603400501976853e-06, 'samples': 27406848, 'steps': 142743, 'loss/train': 1.1290799379348755}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:24 - INFO - __main__ - Step 142749: {'lr': 2.956270222429891e-06, 'samples': 27407808, 'steps': 142748, 'loss/train': 1.3382210731506348}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:27 - INFO - __main__ - Step 142754: {'lr': 2.952203177514379e-06, 'samples': 27408768, 'steps': 142753, 'loss/train': 1.4169700145721436}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:27 - INFO - __main__ - Step 142754: {'lr': 2.952203177514379e-06, 'samples': 27408768, 'steps': 142753, 'loss/train': 1.4169700145721436}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:30 - INFO - __main__ - Step 142760: {'lr': 2.9473263970453134e-06, 'samples': 27409920, 'steps': 142759, 'loss/train': 1.4970378875732422}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:32 - INFO - __main__ - Step 142765: {'lr': 2.943265474566109e-06, 'samples': 27410880, 'steps': 142764, 'loss/train': 0.89985191822052}22}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:35 - INFO - __main__ - Step 142769: {'lr': 2.940018740339695e-06, 'samples': 27411648, 'steps': 142768, 'loss/train': 1.2064054012298584}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:35 - INFO - __main__ - Step 142769: {'lr': 2.940018740339695e-06, 'samples': 27411648, 'steps': 142768, 'loss/train': 1.2064054012298584}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:38 - INFO - __main__ - Step 142776: {'lr': 2.934341241332017e-06, 'samples': 27412992, 'steps': 142775, 'loss/train': 1.7162292003631592}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:40 - INFO - __main__ - Step 142780: {'lr': 2.931099405304144e-06, 'samples': 27413760, 'steps': 142779, 'loss/train': 1.0127644538879395}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:42 - INFO - __main__ - Step 142785: {'lr': 2.927049615094296e-06, 'samples': 27414720, 'steps': 142784, 'loss/train': 1.4091078042984009}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:44 - INFO - __main__ - Step 142789: {'lr': 2.9238117868149173e-06, 'samples': 27415488, 'steps': 142788, 'loss/train': 1.1864008903503418}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:47 - INFO - __main__ - Step 142793: {'lr': 2.920575739795012e-06, 'samples': 27416256, 'steps': 142792, 'loss/train': 1.212540626525879}8}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:48 - INFO - __main__ - Step 142797: {'lr': 2.917341474057894e-06, 'samples': 27417024, 'steps': 142796, 'loss/train': 1.3105329275131226}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:50 - INFO - __main__ - Step 142801: {'lr': 2.9141089896269345e-06, 'samples': 27417792, 'steps': 142800, 'loss/train': 1.2109090089797974}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:50 - INFO - __main__ - Step 142801: {'lr': 2.9141089896269345e-06, 'samples': 27417792, 'steps': 142800, 'loss/train': 1.2109090089797974}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:50 - INFO - __main__ - Step 142801: {'lr': 2.9141089896269345e-06, 'samples': 27417792, 'steps': 142800, 'loss/train': 1.2109090089797974}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:56 - INFO - __main__ - Step 142812: {'lr': 2.905228842491697e-06, 'samples': 27419904, 'steps': 142811, 'loss/train': 1.3414078950881958}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:58 - INFO - __main__ - Step 142817: {'lr': 2.90119686543025e-06, 'samples': 27420864, 'steps': 142816, 'loss/train': 1.6467764377593994}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:20:58 - INFO - __main__ - Step 142817: {'lr': 2.90119686543025e-06, 'samples': 27420864, 'steps': 142816, 'loss/train': 1.6467764377593994}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:03 - INFO - __main__ - Step 142825: {'lr': 2.8947514917739837e-06, 'samples': 27422400, 'steps': 142824, 'loss/train': 1.4187983274459839}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:04 - INFO - __main__ - Step 142829: {'lr': 2.891531477137721e-06, 'samples': 27423168, 'steps': 142828, 'loss/train': 1.096915364265442}9}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:06 - INFO - __main__ - Step 142833: {'lr': 2.888313243993662e-06, 'samples': 27423936, 'steps': 142832, 'loss/train': 0.8896267414093018}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:09 - INFO - __main__ - Step 142838: {'lr': 2.884292957822332e-06, 'samples': 27424896, 'steps': 142837, 'loss/train': 0.7986182570457458}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:09 - INFO - __main__ - Step 142838: {'lr': 2.884292957822332e-06, 'samples': 27424896, 'steps': 142837, 'loss/train': 0.7986182570457458}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:12 - INFO - __main__ - Step 142845: {'lr': 2.8786692337466614e-06, 'samples': 27426240, 'steps': 142844, 'loss/train': 1.3728653192520142}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:14 - INFO - __main__ - Step 142849: {'lr': 2.8754581268033152e-06, 'samples': 27427008, 'steps': 142848, 'loss/train': 1.1242774724960327}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:14 - INFO - __main__ - Step 142849: {'lr': 2.8754581268033152e-06, 'samples': 27427008, 'steps': 142848, 'loss/train': 1.1242774724960327}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:19 - INFO - __main__ - Step 142857: {'lr': 2.8690412577639937e-06, 'samples': 27428544, 'steps': 142856, 'loss/train': 0.3754728138446808}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:20 - INFO - __main__ - Step 142861: {'lr': 2.8658354957142587e-06, 'samples': 27429312, 'steps': 142860, 'loss/train': 1.4112039804458618}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:23 - INFO - __main__ - Step 142866: {'lr': 2.8618307986387484e-06, 'samples': 27430272, 'steps': 142865, 'loss/train': 1.5972285270690918}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:25 - INFO - __main__ - Step 142870: {'lr': 2.8586290453957997e-06, 'samples': 27431040, 'steps': 142869, 'loss/train': 1.685990333557129}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:25 - INFO - __main__ - Step 142870: {'lr': 2.8586290453957997e-06, 'samples': 27431040, 'steps': 142869, 'loss/train': 1.685990333557129}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:28 - INFO - __main__ - Step 142877: {'lr': 2.853030264520912e-06, 'samples': 27432384, 'steps': 142876, 'loss/train': 1.350154161453247}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:30 - INFO - __main__ - Step 142882: {'lr': 2.849034476135659e-06, 'samples': 27433344, 'steps': 142881, 'loss/train': 1.0409126281738281}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:30 - INFO - __main__ - Step 142882: {'lr': 2.849034476135659e-06, 'samples': 27433344, 'steps': 142881, 'loss/train': 1.0409126281738281}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:30 - INFO - __main__ - Step 142882: {'lr': 2.849034476135659e-06, 'samples': 27433344, 'steps': 142881, 'loss/train': 1.0409126281738281}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:36 - INFO - __main__ - Step 142893: {'lr': 2.8402535416436613e-06, 'samples': 27435456, 'steps': 142892, 'loss/train': 0.1609542965888977}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:38 - INFO - __main__ - Step 142897: {'lr': 2.8370638155215123e-06, 'samples': 27436224, 'steps': 142896, 'loss/train': 0.7491282224655151}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:41 - INFO - __main__ - Step 142902: {'lr': 2.83307916364714e-06, 'samples': 27437184, 'steps': 142901, 'loss/train': 1.358563780784607}51}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:43 - INFO - __main__ - Step 142906: {'lr': 2.829893446798293e-06, 'samples': 27437952, 'steps': 142905, 'loss/train': 1.0453078746795654}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:45 - INFO - __main__ - Step 142910: {'lr': 2.8267095118860166e-06, 'samples': 27438720, 'steps': 142909, 'loss/train': 0.6379693746566772}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:46 - INFO - __main__ - Step 142914: {'lr': 2.8235273589332923e-06, 'samples': 27439488, 'steps': 142913, 'loss/train': 1.6284360885620117}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:49 - INFO - __main__ - Step 142918: {'lr': 2.8203469879630182e-06, 'samples': 27440256, 'steps': 142917, 'loss/train': 1.394483208656311}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:51 - INFO - __main__ - Step 142923: {'lr': 2.816374030197966e-06, 'samples': 27441216, 'steps': 142922, 'loss/train': 0.41859015822410583}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:53 - INFO - __main__ - Step 142927: {'lr': 2.813197668772055e-06, 'samples': 27441984, 'steps': 142926, 'loss/train': 1.1697335243225098}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:53 - INFO - __main__ - Step 142927: {'lr': 2.813197668772055e-06, 'samples': 27441984, 'steps': 142926, 'loss/train': 1.1697335243225098}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:57 - INFO - __main__ - Step 142934: {'lr': 2.8076433243650056e-06, 'samples': 27443328, 'steps': 142933, 'loss/train': 1.7718220949172974}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:21:59 - INFO - __main__ - Step 142939: {'lr': 2.8036792769275122e-06, 'samples': 27444288, 'steps': 142938, 'loss/train': 1.5796806812286377}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:01 - INFO - __main__ - Step 142943: {'lr': 2.8005100438666386e-06, 'samples': 27445056, 'steps': 142942, 'loss/train': 1.0324208736419678}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:03 - INFO - __main__ - Step 142947: {'lr': 2.7973425929541663e-06, 'samples': 27445824, 'steps': 142946, 'loss/train': 1.817043423652649}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:03 - INFO - __main__ - Step 142947: {'lr': 2.7973425929541663e-06, 'samples': 27445824, 'steps': 142946, 'loss/train': 1.817043423652649}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:06 - INFO - __main__ - Step 142954: {'lr': 2.791803842220664e-06, 'samples': 27447168, 'steps': 142953, 'loss/train': 1.2838634252548218}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:06 - INFO - __main__ - Step 142954: {'lr': 2.791803842220664e-06, 'samples': 27447168, 'steps': 142953, 'loss/train': 1.2838634252548218}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:11 - INFO - __main__ - Step 142960: {'lr': 2.787060685727011e-06, 'samples': 27448320, 'steps': 142959, 'loss/train': 1.1782252788543701}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:11 - INFO - __main__ - Step 142960: {'lr': 2.787060685727011e-06, 'samples': 27448320, 'steps': 142959, 'loss/train': 1.1782252788543701}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:15 - INFO - __main__ - Step 142969: {'lr': 2.7799534698582372e-06, 'samples': 27450048, 'steps': 142968, 'loss/train': 0.871694028377533}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:17 - INFO - __main__ - Step 142973: {'lr': 2.776797603466469e-06, 'samples': 27450816, 'steps': 142972, 'loss/train': 0.6568059921264648}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:19 - INFO - __main__ - Step 142978: {'lr': 2.7728552768660486e-06, 'samples': 27451776, 'steps': 142977, 'loss/train': 1.2777717113494873}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:21 - INFO - __main__ - Step 142982: {'lr': 2.7697034207248006e-06, 'samples': 27452544, 'steps': 142981, 'loss/train': 1.5090469121932983}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:21 - INFO - __main__ - Step 142982: {'lr': 2.7697034207248006e-06, 'samples': 27452544, 'steps': 142981, 'loss/train': 1.5090469121932983}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:25 - INFO - __main__ - Step 142989: {'lr': 2.764191961320328e-06, 'samples': 27453888, 'steps': 142988, 'loss/train': 0.42998409271240234}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:27 - INFO - __main__ - Step 142994: {'lr': 2.760258546614963e-06, 'samples': 27454848, 'steps': 142993, 'loss/train': 0.8884726762771606}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:27 - INFO - __main__ - Step 142994: {'lr': 2.760258546614963e-06, 'samples': 27454848, 'steps': 142993, 'loss/train': 0.8884726762771606}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:31 - INFO - __main__ - Step 143001: {'lr': 2.7547564449386665e-06, 'samples': 27456192, 'steps': 143000, 'loss/train': 1.212971806526184}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:33 - INFO - __main__ - Step 143005: {'lr': 2.7516148377377813e-06, 'samples': 27456960, 'steps': 143004, 'loss/train': 0.8112437129020691}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:35 - INFO - __main__ - Step 143009: {'lr': 2.7484750130377657e-06, 'samples': 27457728, 'steps': 143008, 'loss/train': 1.033136248588562}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:37 - INFO - __main__ - Step 143014: {'lr': 2.7445527388391676e-06, 'samples': 27458688, 'steps': 143013, 'loss/train': 1.177678108215332}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:39 - INFO - __main__ - Step 143018: {'lr': 2.741416924848972e-06, 'samples': 27459456, 'steps': 143017, 'loss/train': 1.4488118886947632}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:39 - INFO - __main__ - Step 143018: {'lr': 2.741416924848972e-06, 'samples': 27459456, 'steps': 143017, 'loss/train': 1.4488118886947632}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:43 - INFO - __main__ - Step 143025: {'lr': 2.7359335396994202e-06, 'samples': 27460800, 'steps': 143024, 'loss/train': 1.5098410844802856}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:45 - INFO - __main__ - Step 143030: {'lr': 2.732020178415312e-06, 'samples': 27461760, 'steps': 143029, 'loss/train': 1.440919041633606}6}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:48 - INFO - __main__ - Step 143035: {'lr': 2.7281096025096043e-06, 'samples': 27462720, 'steps': 143034, 'loss/train': 1.3328845500946045}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:48 - INFO - __main__ - Step 143035: {'lr': 2.7281096025096043e-06, 'samples': 27462720, 'steps': 143034, 'loss/train': 1.3328845500946045}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:51 - INFO - __main__ - Step 143042: {'lr': 2.7226394757611795e-06, 'samples': 27464064, 'steps': 143041, 'loss/train': 1.6627137660980225}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:53 - INFO - __main__ - Step 143046: {'lr': 2.719516140266054e-06, 'samples': 27464832, 'steps': 143045, 'loss/train': 1.313065767288208}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:55 - INFO - __main__ - Step 143051: {'lr': 2.7156144778670845e-06, 'samples': 27465792, 'steps': 143050, 'loss/train': 0.9041110277175903}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:55 - INFO - __main__ - Step 143051: {'lr': 2.7156144778670845e-06, 'samples': 27465792, 'steps': 143050, 'loss/train': 0.9041110277175903}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:22:59 - INFO - __main__ - Step 143058: {'lr': 2.7101568302645705e-06, 'samples': 27467136, 'steps': 143057, 'loss/train': 1.0545554161071777}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:00 - INFO - __main__ - Step 143062: {'lr': 2.7070406258336845e-06, 'samples': 27467904, 'steps': 143061, 'loss/train': 1.5361019372940063}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:03 - INFO - __main__ - Step 143066: {'lr': 2.703926204225077e-06, 'samples': 27468672, 'steps': 143065, 'loss/train': 1.223467469215393}3}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:05 - INFO - __main__ - Step 143071: {'lr': 2.7000356843423223e-06, 'samples': 27469632, 'steps': 143070, 'loss/train': 1.1606264114379883}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:07 - INFO - __main__ - Step 143075: {'lr': 2.6969252741658736e-06, 'samples': 27470400, 'steps': 143074, 'loss/train': 1.1574069261550903}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:07 - INFO - __main__ - Step 143075: {'lr': 2.6969252741658736e-06, 'samples': 27470400, 'steps': 143074, 'loss/train': 1.1574069261550903}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:11 - INFO - __main__ - Step 143082: {'lr': 2.691486346462102e-06, 'samples': 27471744, 'steps': 143081, 'loss/train': 1.6093335151672363}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:11 - INFO - __main__ - Step 143082: {'lr': 2.691486346462102e-06, 'samples': 27471744, 'steps': 143081, 'loss/train': 1.6093335151672363}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:15 - INFO - __main__ - Step 143091: {'lr': 2.6845014626360187e-06, 'samples': 27473472, 'steps': 143090, 'loss/train': 2.1967554092407227}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:17 - INFO - __main__ - Step 143095: {'lr': 2.6813999671594923e-06, 'samples': 27474240, 'steps': 143094, 'loss/train': 1.1773277521133423}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:19 - INFO - __main__ - Step 143099: {'lr': 2.678300254690097e-06, 'samples': 27475008, 'steps': 143098, 'loss/train': 1.2862331867218018}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:21 - INFO - __main__ - Step 143104: {'lr': 2.674428121491157e-06, 'samples': 27475968, 'steps': 143103, 'loss/train': 1.1739343404769897}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:21 - INFO - __main__ - Step 143104: {'lr': 2.674428121491157e-06, 'samples': 27475968, 'steps': 143103, 'loss/train': 1.1739343404769897}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:21 - INFO - __main__ - Step 143104: {'lr': 2.674428121491157e-06, 'samples': 27475968, 'steps': 143103, 'loss/train': 1.1739343404769897}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:27 - INFO - __main__ - Step 143115: {'lr': 2.665919235330383e-06, 'samples': 27478080, 'steps': 143114, 'loss/train': 0.4351027309894562}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:29 - INFO - __main__ - Step 143119: {'lr': 2.6628284382315125e-06, 'samples': 27478848, 'steps': 143118, 'loss/train': 1.5804588794708252}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:31 - INFO - __main__ - Step 143124: {'lr': 2.6589674494025552e-06, 'samples': 27479808, 'steps': 143123, 'loss/train': 1.2154134511947632}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:34 - INFO - __main__ - Step 143128: {'lr': 2.6558806644022826e-06, 'samples': 27480576, 'steps': 143127, 'loss/train': 1.1656684875488281}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:34 - INFO - __main__ - Step 143128: {'lr': 2.6558806644022826e-06, 'samples': 27480576, 'steps': 143127, 'loss/train': 1.1656684875488281}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:37 - INFO - __main__ - Step 143135: {'lr': 2.650483081468702e-06, 'samples': 27481920, 'steps': 143134, 'loss/train': 1.374346375465393}1}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:39 - INFO - __main__ - Step 143140: {'lr': 2.6466310086368794e-06, 'samples': 27482880, 'steps': 143139, 'loss/train': 1.4214788675308228}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:42 - INFO - __main__ - Step 143145: {'lr': 2.642781722145354e-06, 'samples': 27483840, 'steps': 143144, 'loss/train': 1.4331691265106201}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:44 - INFO - __main__ - Step 143149: {'lr': 2.639704299146284e-06, 'samples': 27484608, 'steps': 143148, 'loss/train': 1.2357772588729858}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:44 - INFO - __main__ - Step 143149: {'lr': 2.639704299146284e-06, 'samples': 27484608, 'steps': 143148, 'loss/train': 1.2357772588729858}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:47 - INFO - __main__ - Step 143156: {'lr': 2.634323099995395e-06, 'samples': 27485952, 'steps': 143155, 'loss/train': 1.6924432516098022}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:49 - INFO - __main__ - Step 143161: {'lr': 2.6304827300839828e-06, 'samples': 27486912, 'steps': 143160, 'loss/train': 1.131056308746338}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:49 - INFO - __main__ - Step 143161: {'lr': 2.6304827300839828e-06, 'samples': 27486912, 'steps': 143160, 'loss/train': 1.131056308746338}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:54 - INFO - __main__ - Step 143169: {'lr': 2.624343934209822e-06, 'samples': 27488448, 'steps': 143168, 'loss/train': 1.316493272781372}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:55 - INFO - __main__ - Step 143173: {'lr': 2.6212772113893834e-06, 'samples': 27489216, 'steps': 143172, 'loss/train': 1.3320523500442505}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:57 - INFO - __main__ - Step 143177: {'lr': 2.6182122720094794e-06, 'samples': 27489984, 'steps': 143176, 'loss/train': 1.459452509880066}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:59 - INFO - __main__ - Step 143182: {'lr': 2.6143836057815616e-06, 'samples': 27490944, 'steps': 143181, 'loss/train': 1.4009207487106323}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:23:59 - INFO - __main__ - Step 143182: {'lr': 2.6143836057815616e-06, 'samples': 27490944, 'steps': 143181, 'loss/train': 1.4009207487106323}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:04 - INFO - __main__ - Step 143190: {'lr': 2.6082635361781848e-06, 'samples': 27492480, 'steps': 143189, 'loss/train': 0.5885257124900818}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:06 - INFO - __main__ - Step 143194: {'lr': 2.605206176667041e-06, 'samples': 27493248, 'steps': 143193, 'loss/train': 1.8182318210601807}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:07 - INFO - __main__ - Step 143198: {'lr': 2.602150600712311e-06, 'samples': 27494016, 'steps': 143197, 'loss/train': 1.3478682041168213}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:10 - INFO - __main__ - Step 143202: {'lr': 2.599096808336032e-06, 'samples': 27494784, 'steps': 143201, 'loss/train': 1.2529464960098267}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:12 - INFO - __main__ - Step 143207: {'lr': 2.5952820760564435e-06, 'samples': 27495744, 'steps': 143206, 'loss/train': 1.271050214767456}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:14 - INFO - __main__ - Step 143211: {'lr': 2.5922322968121583e-06, 'samples': 27496512, 'steps': 143210, 'loss/train': 1.2318823337554932}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:14 - INFO - __main__ - Step 143211: {'lr': 2.5922322968121583e-06, 'samples': 27496512, 'steps': 143210, 'loss/train': 1.2318823337554932}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:17 - INFO - __main__ - Step 143218: {'lr': 2.5868994750556095e-06, 'samples': 27497856, 'steps': 143217, 'loss/train': 1.4821462631225586}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:20 - INFO - __main__ - Step 143223: {'lr': 2.5830936610669597e-06, 'samples': 27498816, 'steps': 143222, 'loss/train': 1.73583984375}5586}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:20 - INFO - __main__ - Step 143223: {'lr': 2.5830936610669597e-06, 'samples': 27498816, 'steps': 143222, 'loss/train': 1.73583984375}5586}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:23 - INFO - __main__ - Step 143230: {'lr': 2.5777702037465267e-06, 'samples': 27500160, 'steps': 143229, 'loss/train': 0.6493167281150818}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:26 - INFO - __main__ - Step 143235: {'lr': 2.5739710787642534e-06, 'samples': 27501120, 'steps': 143234, 'loss/train': 1.334101915359497}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:28 - INFO - __main__ - Step 143240: {'lr': 2.570174740940734e-06, 'samples': 27502080, 'steps': 143239, 'loss/train': 1.092149257659912}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:28 - INFO - __main__ - Step 143240: {'lr': 2.570174740940734e-06, 'samples': 27502080, 'steps': 143239, 'loss/train': 1.092149257659912}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:32 - INFO - __main__ - Step 143247: {'lr': 2.5648645504959265e-06, 'samples': 27503424, 'steps': 143246, 'loss/train': 1.7405369281768799}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:33 - INFO - __main__ - Step 143251: {'lr': 2.561832608738518e-06, 'samples': 27504192, 'steps': 143250, 'loss/train': 1.1998982429504395}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:36 - INFO - __main__ - Step 143255: {'lr': 2.558802450850217e-06, 'samples': 27504960, 'steps': 143254, 'loss/train': 1.446175456047058}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:36 - INFO - __main__ - Step 143255: {'lr': 2.558802450850217e-06, 'samples': 27504960, 'steps': 143254, 'loss/train': 1.446175456047058}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:40 - INFO - __main__ - Step 143263: {'lr': 2.5527474867683697e-06, 'samples': 27506496, 'steps': 143262, 'loss/train': 1.169148564338684}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:41 - INFO - __main__ - Step 143267: {'lr': 2.5497226806184548e-06, 'samples': 27507264, 'steps': 143266, 'loss/train': 1.6762111186981201}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:43 - INFO - __main__ - Step 143271: {'lr': 2.546699658424939e-06, 'samples': 27508032, 'steps': 143270, 'loss/train': 1.0836069583892822}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:46 - INFO - __main__ - Step 143276: {'lr': 2.5429233894050108e-06, 'samples': 27508992, 'steps': 143275, 'loss/train': 0.9185590744018555}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:48 - INFO - __main__ - Step 143281: {'lr': 2.5391499078935845e-06, 'samples': 27509952, 'steps': 143280, 'loss/train': 1.5649242401123047}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:48 - INFO - __main__ - Step 143281: {'lr': 2.5391499078935845e-06, 'samples': 27509952, 'steps': 143280, 'loss/train': 1.5649242401123047}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:52 - INFO - __main__ - Step 143288: {'lr': 2.5338717168727767e-06, 'samples': 27511296, 'steps': 143287, 'loss/train': 1.239648699760437}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:53 - INFO - __main__ - Step 143292: {'lr': 2.5308580608077726e-06, 'samples': 27512064, 'steps': 143291, 'loss/train': 1.306061863899231}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:56 - INFO - __main__ - Step 143297: {'lr': 2.5270934996090287e-06, 'samples': 27513024, 'steps': 143296, 'loss/train': 0.8538229465484619}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:58 - INFO - __main__ - Step 143301: {'lr': 2.524083857782522e-06, 'samples': 27513792, 'steps': 143300, 'loss/train': 1.268620491027832}9}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:24:58 - INFO - __main__ - Step 143301: {'lr': 2.524083857782522e-06, 'samples': 27513792, 'steps': 143300, 'loss/train': 1.268620491027832}9}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:01 - INFO - __main__ - Step 143308: {'lr': 2.5188212776890705e-06, 'samples': 27515136, 'steps': 143307, 'loss/train': 1.5394266843795776}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:03 - INFO - __main__ - Step 143312: {'lr': 2.5158165423032067e-06, 'samples': 27515904, 'steps': 143311, 'loss/train': 1.4295976161956787}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:06 - INFO - __main__ - Step 143317: {'lr': 2.5120631321058907e-06, 'samples': 27516864, 'steps': 143316, 'loss/train': 1.2012109756469727}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:06 - INFO - __main__ - Step 143317: {'lr': 2.5120631321058907e-06, 'samples': 27516864, 'steps': 143316, 'loss/train': 1.2012109756469727}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:10 - INFO - __main__ - Step 143325: {'lr': 2.5060634745484867e-06, 'samples': 27518400, 'steps': 143324, 'loss/train': 1.5147203207015991}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:12 - INFO - __main__ - Step 143329: {'lr': 2.5030663221657646e-06, 'samples': 27519168, 'steps': 143328, 'loss/train': 1.4835432767868042}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:14 - INFO - __main__ - Step 143334: {'lr': 2.499322390851572e-06, 'samples': 27520128, 'steps': 143333, 'loss/train': 1.5615357160568237}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:16 - INFO - __main__ - Step 143338: {'lr': 2.4963292531579584e-06, 'samples': 27520896, 'steps': 143337, 'loss/train': 1.2971636056900024}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:18 - INFO - __main__ - Step 143342: {'lr': 2.493337899805742e-06, 'samples': 27521664, 'steps': 143341, 'loss/train': 1.6153002977371216}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:20 - INFO - __main__ - Step 143346: {'lr': 2.4903483308164054e-06, 'samples': 27522432, 'steps': 143345, 'loss/train': 1.5697529315948486}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:22 - INFO - __main__ - Step 143350: {'lr': 2.487360546211542e-06, 'samples': 27523200, 'steps': 143349, 'loss/train': 1.0795588493347168}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:22 - INFO - __main__ - Step 143350: {'lr': 2.487360546211542e-06, 'samples': 27523200, 'steps': 143349, 'loss/train': 1.0795588493347168}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:26 - INFO - __main__ - Step 143358: {'lr': 2.4813903302414175e-06, 'samples': 27524736, 'steps': 143357, 'loss/train': 1.1240663528442383}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:28 - INFO - __main__ - Step 143362: {'lr': 2.478407898919177e-06, 'samples': 27525504, 'steps': 143361, 'loss/train': 1.778362512588501}3}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:30 - INFO - __main__ - Step 143366: {'lr': 2.4754272520674804e-06, 'samples': 27526272, 'steps': 143365, 'loss/train': 1.4113589525222778}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:32 - INFO - __main__ - Step 143370: {'lr': 2.4724483897078653e-06, 'samples': 27527040, 'steps': 143369, 'loss/train': 0.9012822508811951}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:34 - INFO - __main__ - Step 143375: {'lr': 2.468727321232961e-06, 'samples': 27528000, 'steps': 143374, 'loss/train': 1.1661381721496582}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:36 - INFO - __main__ - Step 143379: {'lr': 2.4657524740589187e-06, 'samples': 27528768, 'steps': 143378, 'loss/train': 1.9400511980056763}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:38 - INFO - __main__ - Step 143383: {'lr': 2.4627794114467086e-06, 'samples': 27529536, 'steps': 143382, 'loss/train': 1.3653523921966553}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:40 - INFO - __main__ - Step 143387: {'lr': 2.4598081334177015e-06, 'samples': 27530304, 'steps': 143386, 'loss/train': 0.9914727210998535}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:42 - INFO - __main__ - Step 143391: {'lr': 2.4568386399933253e-06, 'samples': 27531072, 'steps': 143390, 'loss/train': 1.388185977935791}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:44 - INFO - __main__ - Step 143396: {'lr': 2.453129282845801e-06, 'samples': 27532032, 'steps': 143395, 'loss/train': 1.516230821609497}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:46 - INFO - __main__ - Step 143400: {'lr': 2.450163804860156e-06, 'samples': 27532800, 'steps': 143399, 'loss/train': 1.7190394401550293}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:46 - INFO - __main__ - Step 143400: {'lr': 2.450163804860156e-06, 'samples': 27532800, 'steps': 143399, 'loss/train': 1.7190394401550293}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:50 - INFO - __main__ - Step 143407: {'lr': 2.4449785127703683e-06, 'samples': 27534144, 'steps': 143406, 'loss/train': 1.1680355072021484}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:52 - INFO - __main__ - Step 143412: {'lr': 2.441278079033743e-06, 'samples': 27535104, 'steps': 143411, 'loss/train': 1.9653041362762451}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:52 - INFO - __main__ - Step 143412: {'lr': 2.441278079033743e-06, 'samples': 27535104, 'steps': 143411, 'loss/train': 1.9653041362762451}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:56 - INFO - __main__ - Step 143420: {'lr': 2.435363185471773e-06, 'samples': 27536640, 'steps': 143419, 'loss/train': 1.5172559022903442}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:25:56 - INFO - __main__ - Step 143420: {'lr': 2.435363185471773e-06, 'samples': 27536640, 'steps': 143419, 'loss/train': 1.5172559022903442}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:00 - INFO - __main__ - Step 143427: {'lr': 2.4301935099114436e-06, 'samples': 27537984, 'steps': 143426, 'loss/train': 0.773286759853363}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:00 - INFO - __main__ - Step 143427: {'lr': 2.4301935099114436e-06, 'samples': 27537984, 'steps': 143426, 'loss/train': 0.773286759853363}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:04 - INFO - __main__ - Step 143436: {'lr': 2.4235548158886846e-06, 'samples': 27539712, 'steps': 143435, 'loss/train': 1.5283132791519165}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:07 - INFO - __main__ - Step 143440: {'lr': 2.4206071856048005e-06, 'samples': 27540480, 'steps': 143439, 'loss/train': 1.0464386940002441}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:08 - INFO - __main__ - Step 143444: {'lr': 2.417661340208155e-06, 'samples': 27541248, 'steps': 143443, 'loss/train': 1.188372015953064}1}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:10 - INFO - __main__ - Step 143449: {'lr': 2.4139815434923995e-06, 'samples': 27542208, 'steps': 143448, 'loss/train': 1.1710128784179688}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:12 - INFO - __main__ - Step 143453: {'lr': 2.411039714169727e-06, 'samples': 27542976, 'steps': 143452, 'loss/train': 1.3766335248947144}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:15 - INFO - __main__ - Step 143457: {'lr': 2.4080996698032933e-06, 'samples': 27543744, 'steps': 143456, 'loss/train': 1.1595255136489868}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:17 - INFO - __main__ - Step 143461: {'lr': 2.4051614104143027e-06, 'samples': 27544512, 'steps': 143460, 'loss/train': 1.1854991912841797}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:18 - INFO - __main__ - Step 143465: {'lr': 2.4022249360239057e-06, 'samples': 27545280, 'steps': 143464, 'loss/train': 1.4528214931488037}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:18 - INFO - __main__ - Step 143465: {'lr': 2.4022249360239057e-06, 'samples': 27545280, 'steps': 143464, 'loss/train': 1.4528214931488037}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:18 - INFO - __main__ - Step 143465: {'lr': 2.4022249360239057e-06, 'samples': 27545280, 'steps': 143464, 'loss/train': 1.4528214931488037}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:23 - INFO - __main__ - Step 143475: {'lr': 2.3948915595558007e-06, 'samples': 27547200, 'steps': 143474, 'loss/train': 0.9172769784927368}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:26 - INFO - __main__ - Step 143481: {'lr': 2.390496888871774e-06, 'samples': 27548352, 'steps': 143480, 'loss/train': 1.174222469329834}8}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:26 - INFO - __main__ - Step 143481: {'lr': 2.390496888871774e-06, 'samples': 27548352, 'steps': 143480, 'loss/train': 1.174222469329834}8}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:30 - INFO - __main__ - Step 143488: {'lr': 2.3853748494691853e-06, 'samples': 27549696, 'steps': 143487, 'loss/train': 1.63968026638031}8}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:31 - INFO - __main__ - Step 143492: {'lr': 2.3824504243725452e-06, 'samples': 27550464, 'steps': 143491, 'loss/train': 1.4011118412017822}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:33 - INFO - __main__ - Step 143496: {'lr': 2.3795277844382568e-06, 'samples': 27551232, 'steps': 143495, 'loss/train': 1.4608296155929565}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:36 - INFO - __main__ - Step 143501: {'lr': 2.375876994937076e-06, 'samples': 27552192, 'steps': 143500, 'loss/train': 1.412373661994934}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:36 - INFO - __main__ - Step 143501: {'lr': 2.375876994937076e-06, 'samples': 27552192, 'steps': 143500, 'loss/train': 1.412373661994934}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:39 - INFO - __main__ - Step 143507: {'lr': 2.371499729534321e-06, 'samples': 27553344, 'steps': 143506, 'loss/train': 1.183642864227295}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:42 - INFO - __main__ - Step 143513: {'lr': 2.367126480924492e-06, 'samples': 27554496, 'steps': 143512, 'loss/train': 1.3767234086990356}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:44 - INFO - __main__ - Step 143517: {'lr': 2.3642132134379378e-06, 'samples': 27555264, 'steps': 143516, 'loss/train': 0.4315597116947174}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:46 - INFO - __main__ - Step 143522: {'lr': 2.3605741396516277e-06, 'samples': 27556224, 'steps': 143521, 'loss/train': 1.0394747257232666}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:48 - INFO - __main__ - Step 143526: {'lr': 2.357664889105687e-06, 'samples': 27556992, 'steps': 143525, 'loss/train': 1.1621592044830322}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:50 - INFO - __main__ - Step 143530: {'lr': 2.3547574239007883e-06, 'samples': 27557760, 'steps': 143529, 'loss/train': 1.36100435256958}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:52 - INFO - __main__ - Step 143534: {'lr': 2.351851744057887e-06, 'samples': 27558528, 'steps': 143533, 'loss/train': 1.4655019044876099}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:54 - INFO - __main__ - Step 143538: {'lr': 2.348947849597938e-06, 'samples': 27559296, 'steps': 143537, 'loss/train': 1.0668820142745972}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:56 - INFO - __main__ - Step 143543: {'lr': 2.345320492249642e-06, 'samples': 27560256, 'steps': 143542, 'loss/train': 0.9634972810745239}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:58 - INFO - __main__ - Step 143547: {'lr': 2.3424206149778303e-06, 'samples': 27561024, 'steps': 143546, 'loss/train': 1.4577767848968506}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:26:58 - INFO - __main__ - Step 143547: {'lr': 2.3424206149778303e-06, 'samples': 27561024, 'steps': 143546, 'loss/train': 1.4577767848968506}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:02 - INFO - __main__ - Step 143554: {'lr': 2.337350126006055e-06, 'samples': 27562368, 'steps': 143553, 'loss/train': 1.217521071434021}6}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:04 - INFO - __main__ - Step 143558: {'lr': 2.3344551587746176e-06, 'samples': 27563136, 'steps': 143557, 'loss/train': 0.85478675365448}6}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:06 - INFO - __main__ - Step 143563: {'lr': 2.330838960608872e-06, 'samples': 27564096, 'steps': 143562, 'loss/train': 1.2735806703567505}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:09 - INFO - __main__ - Step 143568: {'lr': 2.3272255523406892e-06, 'samples': 27565056, 'steps': 143567, 'loss/train': 1.4118263721466064}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:09 - INFO - __main__ - Step 143568: {'lr': 2.3272255523406892e-06, 'samples': 27565056, 'steps': 143567, 'loss/train': 1.4118263721466064}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:12 - INFO - __main__ - Step 143575: {'lr': 2.3221714678705876e-06, 'samples': 27566400, 'steps': 143574, 'loss/train': 1.2583914995193481}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:14 - INFO - __main__ - Step 143579: {'lr': 2.3192858747905775e-06, 'samples': 27567168, 'steps': 143578, 'loss/train': 0.7474720478057861}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:16 - INFO - __main__ - Step 143584: {'lr': 2.3156813944679444e-06, 'samples': 27568128, 'steps': 143583, 'loss/train': 1.3409390449523926}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:19 - INFO - __main__ - Step 143589: {'lr': 2.3120797042135712e-06, 'samples': 27569088, 'steps': 143588, 'loss/train': 1.3512591123580933}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:21 - INFO - __main__ - Step 143593: {'lr': 2.3092003608865266e-06, 'samples': 27569856, 'steps': 143592, 'loss/train': 1.2763878107070923}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:21 - INFO - __main__ - Step 143593: {'lr': 2.3092003608865266e-06, 'samples': 27569856, 'steps': 143592, 'loss/train': 1.2763878107070923}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:24 - INFO - __main__ - Step 143600: {'lr': 2.3041658068942984e-06, 'samples': 27571200, 'steps': 143599, 'loss/train': 1.2971009016036987}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:26 - INFO - __main__ - Step 143604: {'lr': 2.3012913742658516e-06, 'samples': 27571968, 'steps': 143603, 'loss/train': 1.220024585723877}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:29 - INFO - __main__ - Step 143609: {'lr': 2.2977008446901092e-06, 'samples': 27572928, 'steps': 143608, 'loss/train': 1.047247052192688}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:29 - INFO - __main__ - Step 143609: {'lr': 2.2977008446901092e-06, 'samples': 27572928, 'steps': 143608, 'loss/train': 1.047247052192688}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:32 - INFO - __main__ - Step 143616: {'lr': 2.2926787909480772e-06, 'samples': 27574272, 'steps': 143615, 'loss/train': 0.6334245800971985}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:34 - INFO - __main__ - Step 143620: {'lr': 2.289811501433675e-06, 'samples': 27575040, 'steps': 143619, 'loss/train': 1.1610758304595947}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:37 - INFO - __main__ - Step 143625: {'lr': 2.2862299008667987e-06, 'samples': 27576000, 'steps': 143624, 'loss/train': 1.2247796058654785}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:37 - INFO - __main__ - Step 143625: {'lr': 2.2862299008667987e-06, 'samples': 27576000, 'steps': 143624, 'loss/train': 1.2247796058654785}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:37 - INFO - __main__ - Step 143625: {'lr': 2.2862299008667987e-06, 'samples': 27576000, 'steps': 143624, 'loss/train': 1.2247796058654785}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:42 - INFO - __main__ - Step 143635: {'lr': 2.2790750709735796e-06, 'samples': 27577920, 'steps': 143634, 'loss/train': 1.2281301021575928}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:45 - INFO - __main__ - Step 143641: {'lr': 2.2747875307398414e-06, 'samples': 27579072, 'steps': 143640, 'loss/train': 1.4904459714889526}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:45 - INFO - __main__ - Step 143641: {'lr': 2.2747875307398414e-06, 'samples': 27579072, 'steps': 143640, 'loss/train': 1.4904459714889526}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:48 - INFO - __main__ - Step 143648: {'lr': 2.2697904792347566e-06, 'samples': 27580416, 'steps': 143647, 'loss/train': 2.874821424484253}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:50 - INFO - __main__ - Step 143653: {'lr': 2.2662245054269616e-06, 'samples': 27581376, 'steps': 143652, 'loss/train': 1.597795009613037}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:53 - INFO - __main__ - Step 143658: {'lr': 2.2626613222445914e-06, 'samples': 27582336, 'steps': 143657, 'loss/train': 1.1246695518493652}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:55 - INFO - __main__ - Step 143662: {'lr': 2.259812784975945e-06, 'samples': 27583104, 'steps': 143661, 'loss/train': 0.7808092832565308}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:57 - INFO - __main__ - Step 143666: {'lr': 2.2569660337538043e-06, 'samples': 27583872, 'steps': 143665, 'loss/train': 1.0987778902053833}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:27:57 - INFO - __main__ - Step 143666: {'lr': 2.2569660337538043e-06, 'samples': 27583872, 'steps': 143665, 'loss/train': 1.0987778902053833}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:00 - INFO - __main__ - Step 143673: {'lr': 2.251988516851128e-06, 'samples': 27585216, 'steps': 143672, 'loss/train': 1.545836091041565}3}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:03 - INFO - __main__ - Step 143678: {'lr': 2.2484364965714433e-06, 'samples': 27586176, 'steps': 143677, 'loss/train': 1.5388952493667603}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:03 - INFO - __main__ - Step 143678: {'lr': 2.2484364965714433e-06, 'samples': 27586176, 'steps': 143677, 'loss/train': 1.5388952493667603}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:06 - INFO - __main__ - Step 143685: {'lr': 2.2434683567761627e-06, 'samples': 27587520, 'steps': 143684, 'loss/train': 1.4527827501296997}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:08 - INFO - __main__ - Step 143689: {'lr': 2.240631875719212e-06, 'samples': 27588288, 'steps': 143688, 'loss/train': 2.1509077548980713}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:11 - INFO - __main__ - Step 143694: {'lr': 2.237088786222724e-06, 'samples': 27589248, 'steps': 143693, 'loss/train': 1.3732889890670776}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:13 - INFO - __main__ - Step 143699: {'lr': 2.2335484876798707e-06, 'samples': 27590208, 'steps': 143698, 'loss/train': 1.0876991748809814}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:13 - INFO - __main__ - Step 143699: {'lr': 2.2335484876798707e-06, 'samples': 27590208, 'steps': 143698, 'loss/train': 1.0876991748809814}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:17 - INFO - __main__ - Step 143706: {'lr': 2.2285967585978507e-06, 'samples': 27591552, 'steps': 143705, 'loss/train': 1.3525655269622803}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:18 - INFO - __main__ - Step 143710: {'lr': 2.2257696552387685e-06, 'samples': 27592320, 'steps': 143709, 'loss/train': 1.4785116910934448}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:21 - INFO - __main__ - Step 143715: {'lr': 2.2222382880152937e-06, 'samples': 27593280, 'steps': 143714, 'loss/train': 1.8921105861663818}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:23 - INFO - __main__ - Step 143720: {'lr': 2.2187097119127362e-06, 'samples': 27594240, 'steps': 143719, 'loss/train': 1.7907893657684326}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:25 - INFO - __main__ - Step 143724: {'lr': 2.215888860664428e-06, 'samples': 27595008, 'steps': 143723, 'loss/train': 1.7844791412353516}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:25 - INFO - __main__ - Step 143724: {'lr': 2.215888860664428e-06, 'samples': 27595008, 'steps': 143723, 'loss/train': 1.7844791412353516}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:28 - INFO - __main__ - Step 143731: {'lr': 2.210956669428471e-06, 'samples': 27596352, 'steps': 143730, 'loss/train': 1.2193893194198608}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:31 - INFO - __main__ - Step 143736: {'lr': 2.2074370251796982e-06, 'samples': 27597312, 'steps': 143735, 'loss/train': 1.2137092351913452}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:33 - INFO - __main__ - Step 143741: {'lr': 2.203920172218571e-06, 'samples': 27598272, 'steps': 143740, 'loss/train': 1.2517330646514893}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:35 - INFO - __main__ - Step 143745: {'lr': 2.20110869960341e-06, 'samples': 27599040, 'steps': 143744, 'loss/train': 1.3538066148757935}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:37 - INFO - __main__ - Step 143749: {'lr': 2.198299013457916e-06, 'samples': 27599808, 'steps': 143748, 'loss/train': 0.7147278189659119}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:39 - INFO - __main__ - Step 143753: {'lr': 2.195491113802406e-06, 'samples': 27600576, 'steps': 143752, 'loss/train': 1.356592059135437}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:41 - INFO - __main__ - Step 143757: {'lr': 2.192685000657113e-06, 'samples': 27601344, 'steps': 143756, 'loss/train': 1.5408519506454468}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:43 - INFO - __main__ - Step 143762: {'lr': 2.1891798715362456e-06, 'samples': 27602304, 'steps': 143761, 'loss/train': 1.28948175907135}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:43 - INFO - __main__ - Step 143762: {'lr': 2.1891798715362456e-06, 'samples': 27602304, 'steps': 143761, 'loss/train': 1.28948175907135}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:47 - INFO - __main__ - Step 143769: {'lr': 2.1842773804846283e-06, 'samples': 27603648, 'steps': 143768, 'loss/train': 0.8560854196548462}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:49 - INFO - __main__ - Step 143773: {'lr': 2.1814784135823217e-06, 'samples': 27604416, 'steps': 143772, 'loss/train': 1.2336177825927734}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:51 - INFO - __main__ - Step 143778: {'lr': 2.1779822173788045e-06, 'samples': 27605376, 'steps': 143777, 'loss/train': 1.370948314666748}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:53 - INFO - __main__ - Step 143782: {'lr': 2.1751872703801024e-06, 'samples': 27606144, 'steps': 143781, 'loss/train': 3.065248727798462}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:56 - INFO - __main__ - Step 143787: {'lr': 2.1716960991199075e-06, 'samples': 27607104, 'steps': 143786, 'loss/train': 0.7629773020744324}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:58 - INFO - __main__ - Step 143791: {'lr': 2.168905172126839e-06, 'samples': 27607872, 'steps': 143790, 'loss/train': 0.3294861614704132}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:28:58 - INFO - __main__ - Step 143791: {'lr': 2.168905172126839e-06, 'samples': 27607872, 'steps': 143790, 'loss/train': 0.3294861614704132}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:01 - INFO - __main__ - Step 143798: {'lr': 2.164025349152443e-06, 'samples': 27609216, 'steps': 143797, 'loss/train': 1.342998743057251}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:03 - INFO - __main__ - Step 143803: {'lr': 2.1605431114392617e-06, 'samples': 27610176, 'steps': 143802, 'loss/train': 1.6185953617095947}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:03 - INFO - __main__ - Step 143803: {'lr': 2.1605431114392617e-06, 'samples': 27610176, 'steps': 143802, 'loss/train': 1.6185953617095947}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:07 - INFO - __main__ - Step 143809: {'lr': 2.1563681113838383e-06, 'samples': 27611328, 'steps': 143808, 'loss/train': 1.3844507932662964}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:09 - INFO - __main__ - Step 143814: {'lr': 2.1528920157173337e-06, 'samples': 27612288, 'steps': 143813, 'loss/train': 1.910151481628418}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:11 - INFO - __main__ - Step 143819: {'lr': 2.1494187119528442e-06, 'samples': 27613248, 'steps': 143818, 'loss/train': 1.5058642625808716}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:11 - INFO - __main__ - Step 143819: {'lr': 2.1494187119528442e-06, 'samples': 27613248, 'steps': 143818, 'loss/train': 1.5058642625808716}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:15 - INFO - __main__ - Step 143826: {'lr': 2.1445607771525545e-06, 'samples': 27614592, 'steps': 143825, 'loss/train': 0.938098132610321}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:17 - INFO - __main__ - Step 143830: {'lr': 2.1417872713590247e-06, 'samples': 27615360, 'steps': 143829, 'loss/train': 1.4186443090438843}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:19 - INFO - __main__ - Step 143834: {'lr': 2.139015552462875e-06, 'samples': 27616128, 'steps': 143833, 'loss/train': 1.253993034362793}3}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:21 - INFO - __main__ - Step 143839: {'lr': 2.1355534166976053e-06, 'samples': 27617088, 'steps': 143838, 'loss/train': 1.297722339630127}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:23 - INFO - __main__ - Step 143843: {'lr': 2.1327857183935927e-06, 'samples': 27617856, 'steps': 143842, 'loss/train': 1.7682024240493774}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:23 - INFO - __main__ - Step 143843: {'lr': 2.1327857183935927e-06, 'samples': 27617856, 'steps': 143842, 'loss/train': 1.7682024240493774}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:27 - INFO - __main__ - Step 143850: {'lr': 2.127946546251508e-06, 'samples': 27619200, 'steps': 143849, 'loss/train': 0.992472231388092}4}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:29 - INFO - __main__ - Step 143855: {'lr': 2.1244933453349325e-06, 'samples': 27620160, 'steps': 143854, 'loss/train': 0.8830081820487976}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:31 - INFO - __main__ - Step 143859: {'lr': 2.121732794999581e-06, 'samples': 27620928, 'steps': 143858, 'loss/train': 1.6305679082870483}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:31 - INFO - __main__ - Step 143859: {'lr': 2.121732794999581e-06, 'samples': 27620928, 'steps': 143858, 'loss/train': 1.6305679082870483}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:35 - INFO - __main__ - Step 143866: {'lr': 2.116906131994417e-06, 'samples': 27622272, 'steps': 143865, 'loss/train': 1.332663893699646}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:37 - INFO - __main__ - Step 143871: {'lr': 2.1134618663248284e-06, 'samples': 27623232, 'steps': 143870, 'loss/train': 0.7977849841117859}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:39 - INFO - __main__ - Step 143876: {'lr': 2.1100203930014826e-06, 'samples': 27624192, 'steps': 143875, 'loss/train': 1.3335516452789307}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:39 - INFO - __main__ - Step 143876: {'lr': 2.1100203930014826e-06, 'samples': 27624192, 'steps': 143875, 'loss/train': 1.3335516452789307}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:43 - INFO - __main__ - Step 143883: {'lr': 2.105207021564276e-06, 'samples': 27625536, 'steps': 143882, 'loss/train': 1.5320667028427124}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:45 - INFO - __main__ - Step 143887: {'lr': 2.102458980939831e-06, 'samples': 27626304, 'steps': 143886, 'loss/train': 0.7858112454414368}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:47 - INFO - __main__ - Step 143891: {'lr': 2.099712727496289e-06, 'samples': 27627072, 'steps': 143890, 'loss/train': 0.8630927205085754}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:49 - INFO - __main__ - Step 143896: {'lr': 2.0962824239451893e-06, 'samples': 27628032, 'steps': 143895, 'loss/train': 1.6517266035079956}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:52 - INFO - __main__ - Step 143900: {'lr': 2.0935401917310527e-06, 'samples': 27628800, 'steps': 143899, 'loss/train': 1.5557652711868286}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:52 - INFO - __main__ - Step 143900: {'lr': 2.0935401917310527e-06, 'samples': 27628800, 'steps': 143899, 'loss/train': 1.5557652711868286}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:55 - INFO - __main__ - Step 143907: {'lr': 2.0887455859269764e-06, 'samples': 27630144, 'steps': 143906, 'loss/train': 1.3416599035263062}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:57 - INFO - __main__ - Step 143912: {'lr': 2.085324218638912e-06, 'samples': 27631104, 'steps': 143911, 'loss/train': 1.3711400032043457}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:29:57 - INFO - __main__ - Step 143912: {'lr': 2.085324218638912e-06, 'samples': 27631104, 'steps': 143911, 'loss/train': 1.3711400032043457}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:01 - INFO - __main__ - Step 143919: {'lr': 2.0805389961183752e-06, 'samples': 27632448, 'steps': 143918, 'loss/train': 1.078730821609497}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:03 - INFO - __main__ - Step 143923: {'lr': 2.077807040833768e-06, 'samples': 27633216, 'steps': 143922, 'loss/train': 0.8315672278404236}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:05 - INFO - __main__ - Step 143928: {'lr': 2.0743946102033672e-06, 'samples': 27634176, 'steps': 143927, 'loss/train': 1.066349744796753}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:05 - INFO - __main__ - Step 143928: {'lr': 2.0743946102033672e-06, 'samples': 27634176, 'steps': 143927, 'loss/train': 1.066349744796753}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:10 - INFO - __main__ - Step 143936: {'lr': 2.0689405302062314e-06, 'samples': 27635712, 'steps': 143935, 'loss/train': 1.334816336631775}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:11 - INFO - __main__ - Step 143940: {'lr': 2.066216171331653e-06, 'samples': 27636480, 'steps': 143939, 'loss/train': 1.439577579498291}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:13 - INFO - __main__ - Step 143944: {'lr': 2.0634935998992966e-06, 'samples': 27637248, 'steps': 143943, 'loss/train': 1.7032158374786377}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:16 - INFO - __main__ - Step 143949: {'lr': 2.0600928992293157e-06, 'samples': 27638208, 'steps': 143948, 'loss/train': 2.1595256328582764}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:18 - INFO - __main__ - Step 143953: {'lr': 2.0573743496136165e-06, 'samples': 27638976, 'steps': 143952, 'loss/train': 1.1854177713394165}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:20 - INFO - __main__ - Step 143957: {'lr': 2.0546575875038664e-06, 'samples': 27639744, 'steps': 143956, 'loss/train': 0.6870977282524109}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:20 - INFO - __main__ - Step 143957: {'lr': 2.0546575875038664e-06, 'samples': 27639744, 'steps': 143956, 'loss/train': 0.6870977282524109}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:23 - INFO - __main__ - Step 143964: {'lr': 2.049907555056851e-06, 'samples': 27641088, 'steps': 143963, 'loss/train': 1.2019859552383423}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:26 - INFO - __main__ - Step 143970: {'lr': 2.0458404558466593e-06, 'samples': 27642240, 'steps': 143969, 'loss/train': 1.467926025390625}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:28 - INFO - __main__ - Step 143974: {'lr': 2.0431312908554823e-06, 'samples': 27643008, 'steps': 143973, 'loss/train': 1.334956169128418}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:30 - INFO - __main__ - Step 143978: {'lr': 2.040423913472922e-06, 'samples': 27643776, 'steps': 143977, 'loss/train': 1.1798903942108154}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:30 - INFO - __main__ - Step 143978: {'lr': 2.040423913472922e-06, 'samples': 27643776, 'steps': 143977, 'loss/train': 1.1798903942108154}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:33 - INFO - __main__ - Step 143985: {'lr': 2.0356903045453954e-06, 'samples': 27645120, 'steps': 143984, 'loss/train': 1.1720072031021118}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:33 - INFO - __main__ - Step 143985: {'lr': 2.0356903045453954e-06, 'samples': 27645120, 'steps': 143984, 'loss/train': 1.1720072031021118}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:38 - INFO - __main__ - Step 143993: {'lr': 2.030287169510675e-06, 'samples': 27646656, 'steps': 143992, 'loss/train': 0.5876056551933289}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:39 - INFO - __main__ - Step 143997: {'lr': 2.0275882835353865e-06, 'samples': 27647424, 'steps': 143996, 'loss/train': 1.3622808456420898}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:41 - INFO - __main__ - Step 144001: {'lr': 2.024891185280792e-06, 'samples': 27648192, 'steps': 144000, 'loss/train': 1.0441641807556152}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:41 - INFO - __main__ - Step 144001: {'lr': 2.024891185280792e-06, 'samples': 27648192, 'steps': 144000, 'loss/train': 1.0441641807556152}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:41 - INFO - __main__ - Step 144001: {'lr': 2.024891185280792e-06, 'samples': 27648192, 'steps': 144000, 'loss/train': 1.0441641807556152}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:47 - INFO - __main__ - Step 144012: {'lr': 2.0174833831740624e-06, 'samples': 27650304, 'steps': 144011, 'loss/train': 1.2375253438949585}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:49 - INFO - __main__ - Step 144017: {'lr': 2.0141206698578163e-06, 'samples': 27651264, 'steps': 144016, 'loss/train': 1.3433767557144165}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:52 - INFO - __main__ - Step 144021: {'lr': 2.0114325104980125e-06, 'samples': 27652032, 'steps': 144020, 'loss/train': 1.3026009798049927}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:54 - INFO - __main__ - Step 144025: {'lr': 2.008746138975337e-06, 'samples': 27652800, 'steps': 144024, 'loss/train': 1.3077946901321411}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:55 - INFO - __main__ - Step 144029: {'lr': 2.006061555309191e-06, 'samples': 27653568, 'steps': 144028, 'loss/train': 0.7535853385925293}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:30:57 - INFO - __main__ - Step 144033: {'lr': 2.0033787595188923e-06, 'samples': 27654336, 'steps': 144032, 'loss/train': 2.016050338745117}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:00 - INFO - __main__ - Step 144038: {'lr': 2.0000277790109267e-06, 'samples': 27655296, 'steps': 144037, 'loss/train': 1.845732569694519}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:00 - INFO - __main__ - Step 144038: {'lr': 2.0000277790109267e-06, 'samples': 27655296, 'steps': 144037, 'loss/train': 1.845732569694519}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:04 - INFO - __main__ - Step 144047: {'lr': 1.9940030540544775e-06, 'samples': 27657024, 'steps': 144046, 'loss/train': 0.939106822013855}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:06 - INFO - __main__ - Step 144051: {'lr': 1.9913283039447727e-06, 'samples': 27657792, 'steps': 144050, 'loss/train': 1.2506053447723389}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:08 - INFO - __main__ - Step 144055: {'lr': 1.988655341817136e-06, 'samples': 27658560, 'steps': 144054, 'loss/train': 1.4581196308135986}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:08 - INFO - __main__ - Step 144055: {'lr': 1.988655341817136e-06, 'samples': 27658560, 'steps': 144054, 'loss/train': 1.4581196308135986}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:12 - INFO - __main__ - Step 144062: {'lr': 1.9839819604836327e-06, 'samples': 27659904, 'steps': 144061, 'loss/train': 1.2836588621139526}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:13 - INFO - __main__ - Step 144066: {'lr': 1.981313915406069e-06, 'samples': 27660672, 'steps': 144065, 'loss/train': 1.4344267845153809}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:16 - INFO - __main__ - Step 144071: {'lr': 1.9779813735127108e-06, 'samples': 27661632, 'steps': 144070, 'loss/train': 1.1863861083984375}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:18 - INFO - __main__ - Step 144076: {'lr': 1.9746516254916803e-06, 'samples': 27662592, 'steps': 144075, 'loss/train': 1.4677149057388306}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:18 - INFO - __main__ - Step 144076: {'lr': 1.9746516254916803e-06, 'samples': 27662592, 'steps': 144075, 'loss/train': 1.4677149057388306}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:22 - INFO - __main__ - Step 144083: {'lr': 1.9699946720392603e-06, 'samples': 27663936, 'steps': 144082, 'loss/train': 1.5365313291549683}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:23 - INFO - __main__ - Step 144087: {'lr': 1.9673360144609466e-06, 'samples': 27664704, 'steps': 144086, 'loss/train': 1.8572328090667725}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:25 - INFO - __main__ - Step 144091: {'lr': 1.964679145037729e-06, 'samples': 27665472, 'steps': 144090, 'loss/train': 1.4549620151519775}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:28 - INFO - __main__ - Step 144096: {'lr': 1.9613605728809046e-06, 'samples': 27666432, 'steps': 144095, 'loss/train': 1.3636845350265503}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:30 - INFO - __main__ - Step 144100: {'lr': 1.958707726876563e-06, 'samples': 27667200, 'steps': 144099, 'loss/train': 1.4368934631347656}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:32 - INFO - __main__ - Step 144104: {'lr': 1.956056669089462e-06, 'samples': 27667968, 'steps': 144103, 'loss/train': 2.1863839626312256}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:34 - INFO - __main__ - Step 144108: {'lr': 1.953407399538781e-06, 'samples': 27668736, 'steps': 144107, 'loss/train': 1.2738503217697144}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:36 - INFO - __main__ - Step 144112: {'lr': 1.950759918243533e-06, 'samples': 27669504, 'steps': 144111, 'loss/train': 1.3946664333343506}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:38 - INFO - __main__ - Step 144117: {'lr': 1.9474530813878013e-06, 'samples': 27670464, 'steps': 144116, 'loss/train': 0.46088120341300964}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:38 - INFO - __main__ - Step 144117: {'lr': 1.9474530813878013e-06, 'samples': 27670464, 'steps': 144116, 'loss/train': 0.46088120341300964}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:41 - INFO - __main__ - Step 144124: {'lr': 1.9428282040814262e-06, 'samples': 27671808, 'steps': 144123, 'loss/train': 0.6077592372894287}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:44 - INFO - __main__ - Step 144128: {'lr': 1.9401878759988113e-06, 'samples': 27672576, 'steps': 144127, 'loss/train': 1.3462728261947632}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:46 - INFO - __main__ - Step 144133: {'lr': 1.936889980766049e-06, 'samples': 27673536, 'steps': 144132, 'loss/train': 1.389816164970398}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:46 - INFO - __main__ - Step 144133: {'lr': 1.936889980766049e-06, 'samples': 27673536, 'steps': 144132, 'loss/train': 1.389816164970398}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:50 - INFO - __main__ - Step 144140: {'lr': 1.932277621931683e-06, 'samples': 27674880, 'steps': 144139, 'loss/train': 1.1475903987884521}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:51 - INFO - __main__ - Step 144144: {'lr': 1.9296444473662356e-06, 'samples': 27675648, 'steps': 144143, 'loss/train': 0.9990177750587463}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:53 - INFO - __main__ - Step 144148: {'lr': 1.9270130612276115e-06, 'samples': 27676416, 'steps': 144147, 'loss/train': 1.511178970336914}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:53 - INFO - __main__ - Step 144148: {'lr': 1.9270130612276115e-06, 'samples': 27676416, 'steps': 144147, 'loss/train': 1.511178970336914}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:58 - INFO - __main__ - Step 144156: {'lr': 1.9217556543065508e-06, 'samples': 27677952, 'steps': 144155, 'loss/train': 1.2927778959274292}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:31:59 - INFO - __main__ - Step 144160: {'lr': 1.9191296335620277e-06, 'samples': 27678720, 'steps': 144159, 'loss/train': 1.6278458833694458}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:02 - INFO - __main__ - Step 144164: {'lr': 1.916505401320101e-06, 'samples': 27679488, 'steps': 144163, 'loss/train': 1.2434659004211426}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:04 - INFO - __main__ - Step 144169: {'lr': 1.9132276261283054e-06, 'samples': 27680448, 'steps': 144168, 'loss/train': 1.6518298387527466}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:04 - INFO - __main__ - Step 144169: {'lr': 1.9132276261283054e-06, 'samples': 27680448, 'steps': 144168, 'loss/train': 1.6518298387527466}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:07 - INFO - __main__ - Step 144176: {'lr': 1.9086434357990235e-06, 'samples': 27681792, 'steps': 144175, 'loss/train': 0.9932098984718323}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:10 - INFO - __main__ - Step 144181: {'lr': 1.9053723677134593e-06, 'samples': 27682752, 'steps': 144180, 'loss/train': 1.9519360065460205}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:10 - INFO - __main__ - Step 144181: {'lr': 1.9053723677134593e-06, 'samples': 27682752, 'steps': 144180, 'loss/train': 1.9519360065460205}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:10 - INFO - __main__ - Step 144181: {'lr': 1.9053723677134593e-06, 'samples': 27682752, 'steps': 144180, 'loss/train': 1.9519360065460205}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:15 - INFO - __main__ - Step 144191: {'lr': 1.8988386156499616e-06, 'samples': 27684672, 'steps': 144190, 'loss/train': 1.2637293338775635}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:18 - INFO - __main__ - Step 144196: {'lr': 1.8955759317456079e-06, 'samples': 27685632, 'steps': 144195, 'loss/train': 1.559923529624939}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:18 - INFO - __main__ - Step 144196: {'lr': 1.8955759317456079e-06, 'samples': 27685632, 'steps': 144195, 'loss/train': 1.559923529624939}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:22 - INFO - __main__ - Step 144204: {'lr': 1.8903614506992628e-06, 'samples': 27687168, 'steps': 144203, 'loss/train': 1.4211711883544922}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:24 - INFO - __main__ - Step 144208: {'lr': 1.8877568932316736e-06, 'samples': 27687936, 'steps': 144207, 'loss/train': 0.948702871799469}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:26 - INFO - __main__ - Step 144212: {'lr': 1.8851541244928328e-06, 'samples': 27688704, 'steps': 144211, 'loss/train': 1.5834769010543823}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:28 - INFO - __main__ - Step 144217: {'lr': 1.8819031789977237e-06, 'samples': 27689664, 'steps': 144216, 'loss/train': 1.2966338396072388}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:28 - INFO - __main__ - Step 144217: {'lr': 1.8819031789977237e-06, 'samples': 27689664, 'steps': 144216, 'loss/train': 1.2966338396072388}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:32 - INFO - __main__ - Step 144223: {'lr': 1.878005733746968e-06, 'samples': 27690816, 'steps': 144222, 'loss/train': 1.631124496459961}8}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:34 - INFO - __main__ - Step 144228: {'lr': 1.8747609372003482e-06, 'samples': 27691776, 'steps': 144227, 'loss/train': 0.8778852224349976}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:34 - INFO - __main__ - Step 144228: {'lr': 1.8747609372003482e-06, 'samples': 27691776, 'steps': 144227, 'loss/train': 0.8778852224349976}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:38 - INFO - __main__ - Step 144236: {'lr': 1.8695750764139707e-06, 'samples': 27693312, 'steps': 144235, 'loss/train': 1.5662468671798706}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:40 - INFO - __main__ - Step 144240: {'lr': 1.8669848293011992e-06, 'samples': 27694080, 'steps': 144239, 'loss/train': 1.1275891065597534}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:42 - INFO - __main__ - Step 144244: {'lr': 1.8643963710669732e-06, 'samples': 27694848, 'steps': 144243, 'loss/train': 0.7743146419525146}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:44 - INFO - __main__ - Step 144249: {'lr': 1.8611633139130334e-06, 'samples': 27695808, 'steps': 144248, 'loss/train': 1.0731548070907593}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:46 - INFO - __main__ - Step 144253: {'lr': 1.8585788807236881e-06, 'samples': 27696576, 'steps': 144252, 'loss/train': 1.1261765956878662}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:46 - INFO - __main__ - Step 144253: {'lr': 1.8585788807236881e-06, 'samples': 27696576, 'steps': 144252, 'loss/train': 1.1261765956878662}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:50 - INFO - __main__ - Step 144260: {'lr': 1.8540604272882789e-06, 'samples': 27697920, 'steps': 144259, 'loss/train': 1.7070385217666626}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:52 - INFO - __main__ - Step 144264: {'lr': 1.8514809137263056e-06, 'samples': 27698688, 'steps': 144263, 'loss/train': 1.3194926977157593}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:54 - INFO - __main__ - Step 144269: {'lr': 1.8482590375436536e-06, 'samples': 27699648, 'steps': 144268, 'loss/train': 0.7892988324165344}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:54 - INFO - __main__ - Step 144269: {'lr': 1.8482590375436536e-06, 'samples': 27699648, 'steps': 144268, 'loss/train': 0.7892988324165344}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:32:58 - INFO - __main__ - Step 144277: {'lr': 1.8431098499603893e-06, 'samples': 27701184, 'steps': 144276, 'loss/train': 1.4231586456298828}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:00 - INFO - __main__ - Step 144281: {'lr': 1.8405379397354593e-06, 'samples': 27701952, 'steps': 144280, 'loss/train': 1.3655062913894653}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:02 - INFO - __main__ - Step 144285: {'lr': 1.8379678185797277e-06, 'samples': 27702720, 'steps': 144284, 'loss/train': 1.3275206089019775}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:02 - INFO - __main__ - Step 144285: {'lr': 1.8379678185797277e-06, 'samples': 27702720, 'steps': 144284, 'loss/train': 1.3275206089019775}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:07 - INFO - __main__ - Step 144293: {'lr': 1.8328329435498836e-06, 'samples': 27704256, 'steps': 144292, 'loss/train': 1.2245502471923828}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:08 - INFO - __main__ - Step 144297: {'lr': 1.8302681897128248e-06, 'samples': 27705024, 'steps': 144296, 'loss/train': 1.9819186925888062}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:10 - INFO - __main__ - Step 144301: {'lr': 1.8277052250189885e-06, 'samples': 27705792, 'steps': 144300, 'loss/train': 0.9730716347694397}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:10 - INFO - __main__ - Step 144301: {'lr': 1.8277052250189885e-06, 'samples': 27705792, 'steps': 144300, 'loss/train': 0.9730716347694397}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:14 - INFO - __main__ - Step 144309: {'lr': 1.8225846631348964e-06, 'samples': 27707328, 'steps': 144308, 'loss/train': 1.7717173099517822}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:17 - INFO - __main__ - Step 144313: {'lr': 1.8200270659815555e-06, 'samples': 27708096, 'steps': 144312, 'loss/train': 0.9855076670646667}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:18 - INFO - __main__ - Step 144317: {'lr': 1.817471258045239e-06, 'samples': 27708864, 'steps': 144316, 'loss/train': 0.42763155698776245}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:20 - INFO - __main__ - Step 144321: {'lr': 1.81491723934446e-06, 'samples': 27709632, 'steps': 144320, 'loss/train': 1.6843920946121216}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:23 - INFO - __main__ - Step 144326: {'lr': 1.811727232108873e-06, 'samples': 27710592, 'steps': 144325, 'loss/train': 1.2168244123458862}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:25 - INFO - __main__ - Step 144330: {'lr': 1.809177239255222e-06, 'samples': 27711360, 'steps': 144329, 'loss/train': 1.6766154766082764}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:25 - INFO - __main__ - Step 144330: {'lr': 1.809177239255222e-06, 'samples': 27711360, 'steps': 144329, 'loss/train': 1.6766154766082764}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:28 - INFO - __main__ - Step 144337: {'lr': 1.8047190572641204e-06, 'samples': 27712704, 'steps': 144336, 'loss/train': 1.6356709003448486}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:31 - INFO - __main__ - Step 144343: {'lr': 1.800902119896447e-06, 'samples': 27713856, 'steps': 144342, 'loss/train': 1.1005154848098755}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:31 - INFO - __main__ - Step 144343: {'lr': 1.800902119896447e-06, 'samples': 27713856, 'steps': 144342, 'loss/train': 1.1005154848098755}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:34 - INFO - __main__ - Step 144350: {'lr': 1.7964541147832692e-06, 'samples': 27715200, 'steps': 144349, 'loss/train': 1.094400405883789}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:36 - INFO - __main__ - Step 144354: {'lr': 1.7939148579765863e-06, 'samples': 27715968, 'steps': 144353, 'loss/train': 1.2742011547088623}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:38 - INFO - __main__ - Step 144359: {'lr': 1.7907433033215736e-06, 'samples': 27716928, 'steps': 144358, 'loss/train': 1.0439954996109009}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:40 - INFO - __main__ - Step 144363: {'lr': 1.7882080727025517e-06, 'samples': 27717696, 'steps': 144362, 'loss/train': 1.4745346307754517}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:43 - INFO - __main__ - Step 144368: {'lr': 1.785041550839972e-06, 'samples': 27718656, 'steps': 144367, 'loss/train': 1.1152318716049194}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:45 - INFO - __main__ - Step 144372: {'lr': 1.7825103465011482e-06, 'samples': 27719424, 'steps': 144371, 'loss/train': 0.6552433967590332}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:45 - INFO - __main__ - Step 144372: {'lr': 1.7825103465011482e-06, 'samples': 27719424, 'steps': 144371, 'loss/train': 0.6552433967590332}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:48 - INFO - __main__ - Step 144379: {'lr': 1.7780850448735342e-06, 'samples': 27720768, 'steps': 144378, 'loss/train': 1.2624056339263916}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:51 - INFO - __main__ - Step 144384: {'lr': 1.7749274704827733e-06, 'samples': 27721728, 'steps': 144383, 'loss/train': 1.5313693284988403}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:53 - INFO - __main__ - Step 144389: {'lr': 1.7717726922516774e-06, 'samples': 27722688, 'steps': 144388, 'loss/train': 0.9945452213287354}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:53 - INFO - __main__ - Step 144389: {'lr': 1.7717726922516774e-06, 'samples': 27722688, 'steps': 144388, 'loss/train': 0.9945452213287354}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:56 - INFO - __main__ - Step 144396: {'lr': 1.767360700344084e-06, 'samples': 27724032, 'steps': 144395, 'loss/train': 1.2735872268676758}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:58 - INFO - __main__ - Step 144400: {'lr': 1.7648420228004446e-06, 'samples': 27724800, 'steps': 144399, 'loss/train': 1.6849546432495117}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:33:58 - INFO - __main__ - Step 144400: {'lr': 1.7648420228004446e-06, 'samples': 27724800, 'steps': 144399, 'loss/train': 1.6849546432495117}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:03 - INFO - __main__ - Step 144408: {'lr': 1.7598100365760483e-06, 'samples': 27726336, 'steps': 144407, 'loss/train': 1.039455771446228}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:05 - INFO - __main__ - Step 144412: {'lr': 1.75729672793154e-06, 'samples': 27727104, 'steps': 144411, 'loss/train': 1.6916885375976562}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:06 - INFO - __main__ - Step 144416: {'lr': 1.754785208956361e-06, 'samples': 27727872, 'steps': 144415, 'loss/train': 1.7920451164245605}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:09 - INFO - __main__ - Step 144421: {'lr': 1.7516483269874317e-06, 'samples': 27728832, 'steps': 144420, 'loss/train': 1.4566465616226196}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:09 - INFO - __main__ - Step 144421: {'lr': 1.7516483269874317e-06, 'samples': 27728832, 'steps': 144420, 'loss/train': 1.4566465616226196}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:13 - INFO - __main__ - Step 144429: {'lr': 1.7466351324093855e-06, 'samples': 27730368, 'steps': 144428, 'loss/train': 1.2913132905960083}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:13 - INFO - __main__ - Step 144429: {'lr': 1.7466351324093855e-06, 'samples': 27730368, 'steps': 144428, 'loss/train': 1.2913132905960083}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:17 - INFO - __main__ - Step 144436: {'lr': 1.7422544597532341e-06, 'samples': 27731712, 'steps': 144435, 'loss/train': 1.5128165483474731}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:19 - INFO - __main__ - Step 144440: {'lr': 1.7397536791736446e-06, 'samples': 27732480, 'steps': 144439, 'loss/train': 1.1945816278457642}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:21 - INFO - __main__ - Step 144444: {'lr': 1.737254688389839e-06, 'samples': 27733248, 'steps': 144443, 'loss/train': 1.4648665189743042}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:22 - INFO - __main__ - Step 144448: {'lr': 1.7347574874198024e-06, 'samples': 27734016, 'steps': 144447, 'loss/train': 1.1585453748703003}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:24 - INFO - __main__ - Step 144452: {'lr': 1.7322620762815766e-06, 'samples': 27734784, 'steps': 144451, 'loss/train': 1.2988146543502808}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:27 - INFO - __main__ - Step 144457: {'lr': 1.729145329337184e-06, 'samples': 27735744, 'steps': 144456, 'loss/train': 0.822123646736145}8}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:27 - INFO - __main__ - Step 144457: {'lr': 1.729145329337184e-06, 'samples': 27735744, 'steps': 144456, 'loss/train': 0.822123646736145}8}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:31 - INFO - __main__ - Step 144464: {'lr': 1.7247865820375352e-06, 'samples': 27737088, 'steps': 144463, 'loss/train': 1.370854139328003}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:32 - INFO - __main__ - Step 144468: {'lr': 1.7222983304062411e-06, 'samples': 27737856, 'steps': 144467, 'loss/train': 1.020707130432129}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:35 - INFO - __main__ - Step 144473: {'lr': 1.7191905329465574e-06, 'samples': 27738816, 'steps': 144472, 'loss/train': 1.282628059387207}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:35 - INFO - __main__ - Step 144473: {'lr': 1.7191905329465574e-06, 'samples': 27738816, 'steps': 144472, 'loss/train': 1.282628059387207}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:39 - INFO - __main__ - Step 144481: {'lr': 1.71422387434364e-06, 'samples': 27740352, 'steps': 144480, 'loss/train': 1.115209698677063}7}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:41 - INFO - __main__ - Step 144485: {'lr': 1.7117432300030044e-06, 'samples': 27741120, 'steps': 144484, 'loss/train': 1.1338306665420532}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:42 - INFO - __main__ - Step 144489: {'lr': 1.7092643756600468e-06, 'samples': 27741888, 'steps': 144488, 'loss/train': 1.3964402675628662}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:42 - INFO - __main__ - Step 144489: {'lr': 1.7092643756600468e-06, 'samples': 27741888, 'steps': 144488, 'loss/train': 1.3964402675628662}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:46 - INFO - __main__ - Step 144496: {'lr': 1.7049306877956473e-06, 'samples': 27743232, 'steps': 144495, 'loss/train': 1.4016375541687012}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:48 - INFO - __main__ - Step 144500: {'lr': 1.7024567560385284e-06, 'samples': 27744000, 'steps': 144499, 'loss/train': 1.167291283607483}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:51 - INFO - __main__ - Step 144505: {'lr': 1.699366858622653e-06, 'samples': 27744960, 'steps': 144504, 'loss/train': 1.1717932224273682}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:53 - INFO - __main__ - Step 144510: {'lr': 1.6962797582176227e-06, 'samples': 27745920, 'steps': 144509, 'loss/train': 1.0314580202102661}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:55 - INFO - __main__ - Step 144514: {'lr': 1.6938120917647792e-06, 'samples': 27746688, 'steps': 144513, 'loss/train': 1.1723084449768066}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:57 - INFO - __main__ - Step 144518: {'lr': 1.6913462154388993e-06, 'samples': 27747456, 'steps': 144517, 'loss/train': 0.9355857968330383}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:34:57 - INFO - __main__ - Step 144518: {'lr': 1.6913462154388993e-06, 'samples': 27747456, 'steps': 144517, 'loss/train': 0.9355857968330383}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:00 - INFO - __main__ - Step 144525: {'lr': 1.6870352394151579e-06, 'samples': 27748800, 'steps': 144524, 'loss/train': 1.453264832496643}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:03 - INFO - __main__ - Step 144531: {'lr': 1.6833444806590992e-06, 'samples': 27749952, 'steps': 144530, 'loss/train': 1.2765672206878662}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:05 - INFO - __main__ - Step 144535: {'lr': 1.680886212571092e-06, 'samples': 27750720, 'steps': 144534, 'loss/train': 1.436996340751648}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:05 - INFO - __main__ - Step 144535: {'lr': 1.680886212571092e-06, 'samples': 27750720, 'steps': 144534, 'loss/train': 1.436996340751648}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:08 - INFO - __main__ - Step 144542: {'lr': 1.6765885511450252e-06, 'samples': 27752064, 'steps': 144541, 'loss/train': 1.2746111154556274}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:11 - INFO - __main__ - Step 144547: {'lr': 1.6735221496989306e-06, 'samples': 27753024, 'steps': 144546, 'loss/train': 1.2218117713928223}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:13 - INFO - __main__ - Step 144551: {'lr': 1.6710710425978381e-06, 'samples': 27753792, 'steps': 144550, 'loss/train': 0.9519199728965759}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:13 - INFO - __main__ - Step 144551: {'lr': 1.6710710425978381e-06, 'samples': 27753792, 'steps': 144550, 'loss/train': 0.9519199728965759}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:17 - INFO - __main__ - Step 144559: {'lr': 1.6661741992860802e-06, 'samples': 27755328, 'steps': 144558, 'loss/train': 1.3848116397857666}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:17 - INFO - __main__ - Step 144559: {'lr': 1.6661741992860802e-06, 'samples': 27755328, 'steps': 144558, 'loss/train': 1.3848116397857666}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:20 - INFO - __main__ - Step 144566: {'lr': 1.6618953358914224e-06, 'samples': 27756672, 'steps': 144565, 'loss/train': 0.822769045829773}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:23 - INFO - __main__ - Step 144572: {'lr': 1.6582321026878122e-06, 'samples': 27757824, 'steps': 144571, 'loss/train': 1.4674463272094727}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:23 - INFO - __main__ - Step 144572: {'lr': 1.6582321026878122e-06, 'samples': 27757824, 'steps': 144571, 'loss/train': 1.4674463272094727}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:27 - INFO - __main__ - Step 144579: {'lr': 1.653963422024174e-06, 'samples': 27759168, 'steps': 144578, 'loss/train': 1.4140580892562866}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:28 - INFO - __main__ - Step 144583: {'lr': 1.6515266377442327e-06, 'samples': 27759936, 'steps': 144582, 'loss/train': 1.1741671562194824}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:30 - INFO - __main__ - Step 144587: {'lr': 1.6490916438961501e-06, 'samples': 27760704, 'steps': 144586, 'loss/train': 1.4192904233932495}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:33 - INFO - __main__ - Step 144592: {'lr': 1.6460504194075466e-06, 'samples': 27761664, 'steps': 144591, 'loss/train': 1.7612751722335815}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:33 - INFO - __main__ - Step 144592: {'lr': 1.6460504194075466e-06, 'samples': 27761664, 'steps': 144591, 'loss/train': 1.7612751722335815}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:37 - INFO - __main__ - Step 144600: {'lr': 1.641190279271787e-06, 'samples': 27763200, 'steps': 144599, 'loss/train': 1.7206830978393555}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:39 - INFO - __main__ - Step 144604: {'lr': 1.6387628949546817e-06, 'samples': 27763968, 'steps': 144603, 'loss/train': 1.1024034023284912}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:41 - INFO - __main__ - Step 144609: {'lr': 1.635731182484529e-06, 'samples': 27764928, 'steps': 144608, 'loss/train': 1.3432189226150513}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:43 - INFO - __main__ - Step 144613: {'lr': 1.6333078268706835e-06, 'samples': 27765696, 'steps': 144612, 'loss/train': 1.4767794609069824}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:45 - INFO - __main__ - Step 144617: {'lr': 1.630886261820036e-06, 'samples': 27766464, 'steps': 144616, 'loss/train': 1.4007188081741333}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:45 - INFO - __main__ - Step 144617: {'lr': 1.630886261820036e-06, 'samples': 27766464, 'steps': 144616, 'loss/train': 1.4007188081741333}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:49 - INFO - __main__ - Step 144624: {'lr': 1.626652831576536e-06, 'samples': 27767808, 'steps': 144623, 'loss/train': 1.3962541818618774}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:51 - INFO - __main__ - Step 144629: {'lr': 1.6236323102216398e-06, 'samples': 27768768, 'steps': 144628, 'loss/train': 0.3423522114753723}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:54 - INFO - __main__ - Step 144634: {'lr': 1.6206145867307397e-06, 'samples': 27769728, 'steps': 144633, 'loss/train': 1.2348554134368896}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:54 - INFO - __main__ - Step 144634: {'lr': 1.6206145867307397e-06, 'samples': 27769728, 'steps': 144633, 'loss/train': 1.2348554134368896}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:57 - INFO - __main__ - Step 144641: {'lr': 1.6163944743196323e-06, 'samples': 27771072, 'steps': 144640, 'loss/train': 0.6227352023124695}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:35:59 - INFO - __main__ - Step 144645: {'lr': 1.6139854436993673e-06, 'samples': 27771840, 'steps': 144644, 'loss/train': 2.028494119644165}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:01 - INFO - __main__ - Step 144650: {'lr': 1.6109766736014109e-06, 'samples': 27772800, 'steps': 144649, 'loss/train': 1.1937143802642822}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:04 - INFO - __main__ - Step 144654: {'lr': 1.608571672086112e-06, 'samples': 27773568, 'steps': 144653, 'loss/train': 1.308449149131775}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:04 - INFO - __main__ - Step 144654: {'lr': 1.608571672086112e-06, 'samples': 27773568, 'steps': 144653, 'loss/train': 1.308449149131775}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:07 - INFO - __main__ - Step 144661: {'lr': 1.604367228416559e-06, 'samples': 27774912, 'steps': 144660, 'loss/train': 1.2541768550872803}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:09 - INFO - __main__ - Step 144666: {'lr': 1.6013674120586297e-06, 'samples': 27775872, 'steps': 144665, 'loss/train': 1.5506134033203125}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:12 - INFO - __main__ - Step 144671: {'lr': 1.5983703938154681e-06, 'samples': 27776832, 'steps': 144670, 'loss/train': 1.3954812288284302}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:12 - INFO - __main__ - Step 144671: {'lr': 1.5983703938154681e-06, 'samples': 27776832, 'steps': 144670, 'loss/train': 1.3954812288284302}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:16 - INFO - __main__ - Step 144678: {'lr': 1.5941792691721302e-06, 'samples': 27778176, 'steps': 144677, 'loss/train': 1.2387032508850098}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:17 - INFO - __main__ - Step 144682: {'lr': 1.591786803210815e-06, 'samples': 27778944, 'steps': 144681, 'loss/train': 1.7761955261230469}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:19 - INFO - __main__ - Step 144686: {'lr': 1.5893961281120416e-06, 'samples': 27779712, 'steps': 144685, 'loss/train': 0.9833414554595947}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:21 - INFO - __main__ - Step 144691: {'lr': 1.5864103026652367e-06, 'samples': 27780672, 'steps': 144690, 'loss/train': 1.025070071220398}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:21 - INFO - __main__ - Step 144691: {'lr': 1.5864103026652367e-06, 'samples': 27780672, 'steps': 144690, 'loss/train': 1.025070071220398}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:25 - INFO - __main__ - Step 144697: {'lr': 1.5828310058659256e-06, 'samples': 27781824, 'steps': 144696, 'loss/train': 1.7446943521499634}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:25 - INFO - __main__ - Step 144697: {'lr': 1.5828310058659256e-06, 'samples': 27781824, 'steps': 144696, 'loss/train': 1.7446943521499634}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:29 - INFO - __main__ - Step 144705: {'lr': 1.5780648783883522e-06, 'samples': 27783360, 'steps': 144704, 'loss/train': 1.163017988204956}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:31 - INFO - __main__ - Step 144709: {'lr': 1.5756845010831412e-06, 'samples': 27784128, 'steps': 144708, 'loss/train': 1.3565864562988281}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:33 - INFO - __main__ - Step 144714: {'lr': 1.5727115480173027e-06, 'samples': 27785088, 'steps': 144713, 'loss/train': 1.2517973184585571}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:36 - INFO - __main__ - Step 144719: {'lr': 1.5697413933890292e-06, 'samples': 27786048, 'steps': 144718, 'loss/train': 0.8487987518310547}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:36 - INFO - __main__ - Step 144719: {'lr': 1.5697413933890292e-06, 'samples': 27786048, 'steps': 144718, 'loss/train': 0.8487987518310547}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:39 - INFO - __main__ - Step 144726: {'lr': 1.5655878783482902e-06, 'samples': 27787392, 'steps': 144725, 'loss/train': 1.3845584392547607}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:41 - INFO - __main__ - Step 144730: {'lr': 1.563216903871928e-06, 'samples': 27788160, 'steps': 144729, 'loss/train': 1.2309081554412842}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:43 - INFO - __main__ - Step 144735: {'lr': 1.5602557044685496e-06, 'samples': 27789120, 'steps': 144734, 'loss/train': 0.9455098509788513}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:45 - INFO - __main__ - Step 144739: {'lr': 1.5578887599203283e-06, 'samples': 27789888, 'steps': 144738, 'loss/train': 1.1396561861038208}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:45 - INFO - __main__ - Step 144739: {'lr': 1.5578887599203283e-06, 'samples': 27789888, 'steps': 144738, 'loss/train': 1.1396561861038208}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:49 - INFO - __main__ - Step 144746: {'lr': 1.5537509168221665e-06, 'samples': 27791232, 'steps': 144745, 'loss/train': 1.4627407789230347}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:51 - INFO - __main__ - Step 144751: {'lr': 1.550798672985243e-06, 'samples': 27792192, 'steps': 144750, 'loss/train': 1.3927375078201294}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:54 - INFO - __main__ - Step 144755: {'lr': 1.548438892966908e-06, 'samples': 27792960, 'steps': 144754, 'loss/train': 1.3005834817886353}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:56 - INFO - __main__ - Step 144759: {'lr': 1.5460809041235591e-06, 'samples': 27793728, 'steps': 144758, 'loss/train': 1.367729902267456}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:56 - INFO - __main__ - Step 144759: {'lr': 1.5460809041235591e-06, 'samples': 27793728, 'steps': 144758, 'loss/train': 1.367729902267456}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:36:59 - INFO - __main__ - Step 144766: {'lr': 1.541958733713822e-06, 'samples': 27795072, 'steps': 144765, 'loss/train': 1.2658292055130005}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:01 - INFO - __main__ - Step 144771: {'lr': 1.5390176848137371e-06, 'samples': 27796032, 'steps': 144770, 'loss/train': 0.11231900006532669}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:04 - INFO - __main__ - Step 144775: {'lr': 1.5366668608404188e-06, 'samples': 27796800, 'steps': 144774, 'loss/train': 1.479247808456421}9}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:04 - INFO - __main__ - Step 144775: {'lr': 1.5366668608404188e-06, 'samples': 27796800, 'steps': 144774, 'loss/train': 1.479247808456421}9}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:07 - INFO - __main__ - Step 144782: {'lr': 1.5325572291164102e-06, 'samples': 27798144, 'steps': 144781, 'loss/train': 1.218016266822815}9}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:10 - INFO - __main__ - Step 144787: {'lr': 1.5296251365475855e-06, 'samples': 27799104, 'steps': 144786, 'loss/train': 1.3177590370178223}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:12 - INFO - __main__ - Step 144791: {'lr': 1.5272814777154142e-06, 'samples': 27799872, 'steps': 144790, 'loss/train': 1.0483356714248657}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:14 - INFO - __main__ - Step 144795: {'lr': 1.5249396102108294e-06, 'samples': 27800640, 'steps': 144794, 'loss/train': 1.5502421855926514}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:16 - INFO - __main__ - Step 144799: {'lr': 1.5225995340506782e-06, 'samples': 27801408, 'steps': 144798, 'loss/train': 1.0677552223205566}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:17 - INFO - __main__ - Step 144803: {'lr': 1.5202612492518365e-06, 'samples': 27802176, 'steps': 144802, 'loss/train': 1.2964812517166138}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:20 - INFO - __main__ - Step 144808: {'lr': 1.5173409123808434e-06, 'samples': 27803136, 'steps': 144807, 'loss/train': 1.2971835136413574}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:22 - INFO - __main__ - Step 144812: {'lr': 1.5150066582065857e-06, 'samples': 27803904, 'steps': 144811, 'loss/train': 1.370383858680725}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:24 - INFO - __main__ - Step 144816: {'lr': 1.5126741954483714e-06, 'samples': 27804672, 'steps': 144815, 'loss/train': 1.4596760272979736}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:26 - INFO - __main__ - Step 144820: {'lr': 1.5103435241230757e-06, 'samples': 27805440, 'steps': 144819, 'loss/train': 1.2631117105484009}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:27 - INFO - __main__ - Step 144824: {'lr': 1.5080146442474075e-06, 'samples': 27806208, 'steps': 144823, 'loss/train': 1.5040192604064941}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:30 - INFO - __main__ - Step 144828: {'lr': 1.5056875558382144e-06, 'samples': 27806976, 'steps': 144827, 'loss/train': 0.9846555590629578}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:30 - INFO - __main__ - Step 144828: {'lr': 1.5056875558382144e-06, 'samples': 27806976, 'steps': 144827, 'loss/train': 0.9846555590629578}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:34 - INFO - __main__ - Step 144836: {'lr': 1.5010387534863667e-06, 'samples': 27808512, 'steps': 144835, 'loss/train': 1.3354222774505615}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:35 - INFO - __main__ - Step 144840: {'lr': 1.4987170395771854e-06, 'samples': 27809280, 'steps': 144839, 'loss/train': 1.404546856880188}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:37 - INFO - __main__ - Step 144844: {'lr': 1.4963971172014812e-06, 'samples': 27810048, 'steps': 144843, 'loss/train': 0.5878109931945801}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:40 - INFO - __main__ - Step 144849: {'lr': 1.4934997336013557e-06, 'samples': 27811008, 'steps': 144848, 'loss/train': 1.2410914897918701}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:40 - INFO - __main__ - Step 144849: {'lr': 1.4934997336013557e-06, 'samples': 27811008, 'steps': 144848, 'loss/train': 1.2410914897918701}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:43 - INFO - __main__ - Step 144856: {'lr': 1.4894480994425973e-06, 'samples': 27812352, 'steps': 144855, 'loss/train': 0.8802388906478882}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:46 - INFO - __main__ - Step 144860: {'lr': 1.487135343368029e-06, 'samples': 27813120, 'steps': 144859, 'loss/train': 1.2057926654815674}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:46 - INFO - __main__ - Step 144860: {'lr': 1.487135343368029e-06, 'samples': 27813120, 'steps': 144859, 'loss/train': 1.2057926654815674}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:50 - INFO - __main__ - Step 144868: {'lr': 1.4825152060864989e-06, 'samples': 27814656, 'steps': 144867, 'loss/train': 2.439152240753174}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:52 - INFO - __main__ - Step 144872: {'lr': 1.4802078249128714e-06, 'samples': 27815424, 'steps': 144871, 'loss/train': 0.13792888820171356}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:54 - INFO - __main__ - Step 144876: {'lr': 1.47790223540617e-06, 'samples': 27816192, 'steps': 144875, 'loss/train': 1.1569918394088745}56}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:56 - INFO - __main__ - Step 144881: {'lr': 1.4750227680797313e-06, 'samples': 27817152, 'steps': 144880, 'loss/train': 1.2910583019256592}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:58 - INFO - __main__ - Step 144885: {'lr': 1.47272120988437e-06, 'samples': 27817920, 'steps': 144884, 'loss/train': 1.1047155857086182}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:37:58 - INFO - __main__ - Step 144885: {'lr': 1.47272120988437e-06, 'samples': 27817920, 'steps': 144884, 'loss/train': 1.1047155857086182}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:02 - INFO - __main__ - Step 144892: {'lr': 1.468697794380941e-06, 'samples': 27819264, 'steps': 144891, 'loss/train': 1.3503011465072632}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:04 - INFO - __main__ - Step 144897: {'lr': 1.4658272856900856e-06, 'samples': 27820224, 'steps': 144896, 'loss/train': 0.8283629417419434}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:06 - INFO - __main__ - Step 144902: {'lr': 1.4629595766408322e-06, 'samples': 27821184, 'steps': 144901, 'loss/train': 1.1349883079528809}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:06 - INFO - __main__ - Step 144902: {'lr': 1.4629595766408322e-06, 'samples': 27821184, 'steps': 144901, 'loss/train': 1.1349883079528809}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:10 - INFO - __main__ - Step 144909: {'lr': 1.4589494874311937e-06, 'samples': 27822528, 'steps': 144908, 'loss/train': 0.7284704446792603}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:12 - INFO - __main__ - Step 144913: {'lr': 1.456660471629928e-06, 'samples': 27823296, 'steps': 144912, 'loss/train': 1.3444725275039673}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:14 - INFO - __main__ - Step 144917: {'lr': 1.4543732476653692e-06, 'samples': 27824064, 'steps': 144916, 'loss/train': 1.43563973903656}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:14 - INFO - __main__ - Step 144917: {'lr': 1.4543732476653692e-06, 'samples': 27824064, 'steps': 144916, 'loss/train': 1.43563973903656}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:18 - INFO - __main__ - Step 144925: {'lr': 1.449804175312236e-06, 'samples': 27825600, 'steps': 144924, 'loss/train': 1.328940987586975}4}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:19 - INFO - __main__ - Step 144929: {'lr': 1.447522326956663e-06, 'samples': 27826368, 'steps': 144928, 'loss/train': 1.3839906454086304}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:21 - INFO - __main__ - Step 144933: {'lr': 1.4452422705036328e-06, 'samples': 27827136, 'steps': 144932, 'loss/train': 1.3972604274749756}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:24 - INFO - __main__ - Step 144938: {'lr': 1.4423947198253796e-06, 'samples': 27828096, 'steps': 144937, 'loss/train': 1.2136224508285522}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:24 - INFO - __main__ - Step 144938: {'lr': 1.4423947198253796e-06, 'samples': 27828096, 'steps': 144937, 'loss/train': 1.2136224508285522}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:28 - INFO - __main__ - Step 144946: {'lr': 1.4378444625570418e-06, 'samples': 27829632, 'steps': 144945, 'loss/train': 1.0826929807662964}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:30 - INFO - __main__ - Step 144950: {'lr': 1.4355720218731939e-06, 'samples': 27830400, 'steps': 144949, 'loss/train': 1.0677618980407715}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:32 - INFO - __main__ - Step 144954: {'lr': 1.4333013731780697e-06, 'samples': 27831168, 'steps': 144953, 'loss/train': 1.0710880756378174}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:34 - INFO - __main__ - Step 144959: {'lr': 1.4304655823182688e-06, 'samples': 27832128, 'steps': 144958, 'loss/train': 1.6557073593139648}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:36 - INFO - __main__ - Step 144963: {'lr': 1.4281989656576011e-06, 'samples': 27832896, 'steps': 144962, 'loss/train': 1.1288398504257202}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:38 - INFO - __main__ - Step 144967: {'lr': 1.4259341410388649e-06, 'samples': 27833664, 'steps': 144966, 'loss/train': 1.6161425113677979}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:40 - INFO - __main__ - Step 144971: {'lr': 1.423671108478297e-06, 'samples': 27834432, 'steps': 144970, 'loss/train': 1.480186939239502}9}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:42 - INFO - __main__ - Step 144975: {'lr': 1.4214098679922739e-06, 'samples': 27835200, 'steps': 144974, 'loss/train': 1.4158785343170166}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:44 - INFO - __main__ - Step 144979: {'lr': 1.4191504195970872e-06, 'samples': 27835968, 'steps': 144978, 'loss/train': 2.163895845413208}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:46 - INFO - __main__ - Step 144983: {'lr': 1.4168927633090023e-06, 'samples': 27836736, 'steps': 144982, 'loss/train': 1.4289168119430542}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:48 - INFO - __main__ - Step 144987: {'lr': 1.4146368991443392e-06, 'samples': 27837504, 'steps': 144986, 'loss/train': 1.3249605894088745}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:50 - INFO - __main__ - Step 144991: {'lr': 1.4123828271193072e-06, 'samples': 27838272, 'steps': 144990, 'loss/train': 1.33355712890625}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:52 - INFO - __main__ - Step 144996: {'lr': 1.409567757309188e-06, 'samples': 27839232, 'steps': 144995, 'loss/train': 0.4522007703781128}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:52 - INFO - __main__ - Step 144996: {'lr': 1.409567757309188e-06, 'samples': 27839232, 'steps': 144995, 'loss/train': 0.4522007703781128}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:56 - INFO - __main__ - Step 145003: {'lr': 1.405631364044646e-06, 'samples': 27840576, 'steps': 145002, 'loss/train': 1.0644632577896118}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:38:58 - INFO - __main__ - Step 145008: {'lr': 1.4028230149484844e-06, 'samples': 27841536, 'steps': 145007, 'loss/train': 0.6648860573768616}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:00 - INFO - __main__ - Step 145012: {'lr': 1.4005783519229763e-06, 'samples': 27842304, 'steps': 145011, 'loss/train': 1.3725347518920898}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:00 - INFO - __main__ - Step 145012: {'lr': 1.4005783519229763e-06, 'samples': 27842304, 'steps': 145011, 'loss/train': 1.3725347518920898}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:05 - INFO - __main__ - Step 145020: {'lr': 1.396094402611181e-06, 'samples': 27843840, 'steps': 145019, 'loss/train': 1.710268497467041}8}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:07 - INFO - __main__ - Step 145024: {'lr': 1.3938551163572011e-06, 'samples': 27844608, 'steps': 145023, 'loss/train': 1.5254946947097778}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:08 - INFO - __main__ - Step 145028: {'lr': 1.3916176223927047e-06, 'samples': 27845376, 'steps': 145027, 'loss/train': 0.3621349334716797}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:10 - INFO - __main__ - Step 145033: {'lr': 1.388823275368739e-06, 'samples': 27846336, 'steps': 145032, 'loss/train': 1.1731802225112915}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:13 - INFO - __main__ - Step 145038: {'lr': 1.3860317288537927e-06, 'samples': 27847296, 'steps': 145037, 'loss/train': 1.4008498191833496}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:13 - INFO - __main__ - Step 145038: {'lr': 1.3860317288537927e-06, 'samples': 27847296, 'steps': 145037, 'loss/train': 1.4008498191833496}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:16 - INFO - __main__ - Step 145045: {'lr': 1.3821282686478088e-06, 'samples': 27848640, 'steps': 145044, 'loss/train': 0.8759263157844543}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:18 - INFO - __main__ - Step 145049: {'lr': 1.3799001844674308e-06, 'samples': 27849408, 'steps': 145048, 'loss/train': 1.2305889129638672}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:21 - INFO - __main__ - Step 145054: {'lr': 1.3771175997924213e-06, 'samples': 27850368, 'steps': 145053, 'loss/train': 1.2829804420471191}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:23 - INFO - __main__ - Step 145058: {'lr': 1.3748935485123072e-06, 'samples': 27851136, 'steps': 145057, 'loss/train': 1.0570883750915527}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:25 - INFO - __main__ - Step 145062: {'lr': 1.3726712896584003e-06, 'samples': 27851904, 'steps': 145061, 'loss/train': 1.419749140739441}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:26 - INFO - __main__ - Step 145066: {'lr': 1.3704508232466882e-06, 'samples': 27852672, 'steps': 145065, 'loss/train': 1.1051236391067505}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:28 - INFO - __main__ - Step 145070: {'lr': 1.368232149293186e-06, 'samples': 27853440, 'steps': 145069, 'loss/train': 0.6991701722145081}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:31 - INFO - __main__ - Step 145075: {'lr': 1.3654613275200235e-06, 'samples': 27854400, 'steps': 145074, 'loss/train': 0.928056001663208}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:33 - INFO - __main__ - Step 145079: {'lr': 1.363246686655939e-06, 'samples': 27855168, 'steps': 145078, 'loss/train': 1.3207581043243408}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:33 - INFO - __main__ - Step 145079: {'lr': 1.363246686655939e-06, 'samples': 27855168, 'steps': 145078, 'loss/train': 1.3207581043243408}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:36 - INFO - __main__ - Step 145086: {'lr': 1.3593753783809614e-06, 'samples': 27856512, 'steps': 145085, 'loss/train': 1.3697439432144165}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:39 - INFO - __main__ - Step 145091: {'lr': 1.3566135191884488e-06, 'samples': 27857472, 'steps': 145090, 'loss/train': 1.3762223720550537}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:39 - INFO - __main__ - Step 145091: {'lr': 1.3566135191884488e-06, 'samples': 27857472, 'steps': 145090, 'loss/train': 1.3762223720550537}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:43 - INFO - __main__ - Step 145099: {'lr': 1.3522003703066998e-06, 'samples': 27859008, 'steps': 145098, 'loss/train': 1.986587405204773}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:45 - INFO - __main__ - Step 145103: {'lr': 1.3499964847425717e-06, 'samples': 27859776, 'steps': 145102, 'loss/train': 1.4563846588134766}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:46 - INFO - __main__ - Step 145107: {'lr': 1.3477943917840907e-06, 'samples': 27860544, 'steps': 145106, 'loss/train': 1.1496927738189697}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:48 - INFO - __main__ - Step 145111: {'lr': 1.3455940914471886e-06, 'samples': 27861312, 'steps': 145110, 'loss/train': 1.0594466924667358}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:48 - INFO - __main__ - Step 145111: {'lr': 1.3455940914471886e-06, 'samples': 27861312, 'steps': 145110, 'loss/train': 1.0594466924667358}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:53 - INFO - __main__ - Step 145119: {'lr': 1.3411988687014598e-06, 'samples': 27862848, 'steps': 145118, 'loss/train': 1.7607797384262085}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:54 - INFO - __main__ - Step 145123: {'lr': 1.339003946324302e-06, 'samples': 27863616, 'steps': 145122, 'loss/train': 1.4044004678726196}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:56 - INFO - __main__ - Step 145127: {'lr': 1.3368108166320891e-06, 'samples': 27864384, 'steps': 145126, 'loss/train': 0.49436846375465393}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:39:56 - INFO - __main__ - Step 145127: {'lr': 1.3368108166320891e-06, 'samples': 27864384, 'steps': 145126, 'loss/train': 0.49436846375465393}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:00 - INFO - __main__ - Step 145135: {'lr': 1.332429935365642e-06, 'samples': 27865920, 'steps': 145134, 'loss/train': 2.098557949066162}93}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:03 - INFO - __main__ - Step 145139: {'lr': 1.3302421838230217e-06, 'samples': 27866688, 'steps': 145138, 'loss/train': 1.3355449438095093}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:03 - INFO - __main__ - Step 145139: {'lr': 1.3302421838230217e-06, 'samples': 27866688, 'steps': 145138, 'loss/train': 1.3355449438095093}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:07 - INFO - __main__ - Step 145147: {'lr': 1.3258720589977846e-06, 'samples': 27868224, 'steps': 145146, 'loss/train': 1.002426266670227}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:08 - INFO - __main__ - Step 145151: {'lr': 1.3236896857466706e-06, 'samples': 27868992, 'steps': 145150, 'loss/train': 1.4182416200637817}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:10 - INFO - __main__ - Step 145155: {'lr': 1.3215091052909133e-06, 'samples': 27869760, 'steps': 145154, 'loss/train': 1.3644630908966064}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:13 - INFO - __main__ - Step 145160: {'lr': 1.318785900863567e-06, 'samples': 27870720, 'steps': 145159, 'loss/train': 0.8748659491539001}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:15 - INFO - __main__ - Step 145165: {'lr': 1.3160654977341646e-06, 'samples': 27871680, 'steps': 145164, 'loss/train': 1.772456169128418}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:15 - INFO - __main__ - Step 145165: {'lr': 1.3160654977341646e-06, 'samples': 27871680, 'steps': 145164, 'loss/train': 1.772456169128418}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:18 - INFO - __main__ - Step 145172: {'lr': 1.3122616395918884e-06, 'samples': 27873024, 'steps': 145171, 'loss/train': 1.0145297050476074}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:20 - INFO - __main__ - Step 145176: {'lr': 1.3100904715692142e-06, 'samples': 27873792, 'steps': 145175, 'loss/train': 1.2298520803451538}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:23 - INFO - __main__ - Step 145181: {'lr': 1.3073790327989832e-06, 'samples': 27874752, 'steps': 145180, 'loss/train': 1.3083131313323975}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:23 - INFO - __main__ - Step 145181: {'lr': 1.3073790327989832e-06, 'samples': 27874752, 'steps': 145180, 'loss/train': 1.3083131313323975}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:27 - INFO - __main__ - Step 145189: {'lr': 1.3030465577463236e-06, 'samples': 27876288, 'steps': 145188, 'loss/train': 1.3655767440795898}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:29 - INFO - __main__ - Step 145193: {'lr': 1.3008830096284118e-06, 'samples': 27877056, 'steps': 145192, 'loss/train': 1.4844951629638672}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:31 - INFO - __main__ - Step 145197: {'lr': 1.298721254470281e-06, 'samples': 27877824, 'steps': 145196, 'loss/train': 0.7123776078224182}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:33 - INFO - __main__ - Step 145201: {'lr': 1.296561292287446e-06, 'samples': 27878592, 'steps': 145200, 'loss/train': 1.2082818746566772}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:35 - INFO - __main__ - Step 145206: {'lr': 1.2938638609542031e-06, 'samples': 27879552, 'steps': 145205, 'loss/train': 1.259667158126831}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:37 - INFO - __main__ - Step 145210: {'lr': 1.291707933022801e-06, 'samples': 27880320, 'steps': 145209, 'loss/train': 1.111890196800232}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:39 - INFO - __main__ - Step 145214: {'lr': 1.2895537981172933e-06, 'samples': 27881088, 'steps': 145213, 'loss/train': 1.2562322616577148}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:41 - INFO - __main__ - Step 145218: {'lr': 1.2874014562531954e-06, 'samples': 27881856, 'steps': 145217, 'loss/train': 1.3481478691101074}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:43 - INFO - __main__ - Step 145222: {'lr': 1.2852509074460784e-06, 'samples': 27882624, 'steps': 145221, 'loss/train': 1.0362510681152344}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:45 - INFO - __main__ - Step 145226: {'lr': 1.283102151711374e-06, 'samples': 27883392, 'steps': 145225, 'loss/train': 1.019194483757019}4}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:47 - INFO - __main__ - Step 145230: {'lr': 1.2809551890646253e-06, 'samples': 27884160, 'steps': 145229, 'loss/train': 1.2752697467803955}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:49 - INFO - __main__ - Step 145234: {'lr': 1.2788100195212649e-06, 'samples': 27884928, 'steps': 145233, 'loss/train': 1.1175587177276611}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:51 - INFO - __main__ - Step 145239: {'lr': 1.276131079167353e-06, 'samples': 27885888, 'steps': 145238, 'loss/train': 1.5251662731170654}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:53 - INFO - __main__ - Step 145243: {'lr': 1.273989944163223e-06, 'samples': 27886656, 'steps': 145242, 'loss/train': 1.1150532960891724}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:55 - INFO - __main__ - Step 145247: {'lr': 1.2718506023127464e-06, 'samples': 27887424, 'steps': 145246, 'loss/train': 1.034528136253357}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:55 - INFO - __main__ - Step 145247: {'lr': 1.2718506023127464e-06, 'samples': 27887424, 'steps': 145246, 'loss/train': 1.034528136253357}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:40:59 - INFO - __main__ - Step 145254: {'lr': 1.2681110688966823e-06, 'samples': 27888768, 'steps': 145253, 'loss/train': 1.4038711786270142}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:02 - INFO - __main__ - Step 145260: {'lr': 1.26491012545224e-06, 'samples': 27889920, 'steps': 145259, 'loss/train': 1.6079399585723877}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:04 - INFO - __main__ - Step 145264: {'lr': 1.26277840467659e-06, 'samples': 27890688, 'steps': 145263, 'loss/train': 1.2332956790924072}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:04 - INFO - __main__ - Step 145264: {'lr': 1.26277840467659e-06, 'samples': 27890688, 'steps': 145263, 'loss/train': 1.2332956790924072}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:07 - INFO - __main__ - Step 145271: {'lr': 1.259052208298861e-06, 'samples': 27892032, 'steps': 145270, 'loss/train': 0.1713971346616745}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:10 - INFO - __main__ - Step 145276: {'lr': 1.2563940018178176e-06, 'samples': 27892992, 'steps': 145275, 'loss/train': 1.3292726278305054}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:10 - INFO - __main__ - Step 145276: {'lr': 1.2563940018178176e-06, 'samples': 27892992, 'steps': 145275, 'loss/train': 1.3292726278305054}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:13 - INFO - __main__ - Step 145283: {'lr': 1.2526772201123249e-06, 'samples': 27894336, 'steps': 145282, 'loss/train': 0.9604284763336182}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:15 - INFO - __main__ - Step 145287: {'lr': 1.250555810644316e-06, 'samples': 27895104, 'steps': 145286, 'loss/train': 0.801484227180481}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:15 - INFO - __main__ - Step 145287: {'lr': 1.250555810644316e-06, 'samples': 27895104, 'steps': 145286, 'loss/train': 0.801484227180481}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:19 - INFO - __main__ - Step 145295: {'lr': 1.246318371691285e-06, 'samples': 27896640, 'steps': 145294, 'loss/train': 1.5075080394744873}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:21 - INFO - __main__ - Step 145299: {'lr': 1.244202342236822e-06, 'samples': 27897408, 'steps': 145298, 'loss/train': 1.1255440711975098}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:21 - INFO - __main__ - Step 145299: {'lr': 1.244202342236822e-06, 'samples': 27897408, 'steps': 145298, 'loss/train': 1.1255440711975098}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:21 - INFO - __main__ - Step 145299: {'lr': 1.244202342236822e-06, 'samples': 27897408, 'steps': 145298, 'loss/train': 1.1255440711975098}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:26 - INFO - __main__ - Step 145309: {'lr': 1.2389201146206784e-06, 'samples': 27899328, 'steps': 145308, 'loss/train': 1.210412621498108}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:29 - INFO - __main__ - Step 145314: {'lr': 1.2362832040941075e-06, 'samples': 27900288, 'steps': 145313, 'loss/train': 1.7078617811203003}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:31 - INFO - __main__ - Step 145319: {'lr': 1.2336490957948477e-06, 'samples': 27901248, 'steps': 145318, 'loss/train': 1.3163930177688599}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:31 - INFO - __main__ - Step 145319: {'lr': 1.2336490957948477e-06, 'samples': 27901248, 'steps': 145318, 'loss/train': 1.3163930177688599}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:34 - INFO - __main__ - Step 145326: {'lr': 1.229966051974285e-06, 'samples': 27902592, 'steps': 145325, 'loss/train': 1.487215280532837}9}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:37 - INFO - __main__ - Step 145330: {'lr': 1.2278639215227983e-06, 'samples': 27903360, 'steps': 145329, 'loss/train': 1.262801170349121}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:37 - INFO - __main__ - Step 145330: {'lr': 1.2278639215227983e-06, 'samples': 27903360, 'steps': 145329, 'loss/train': 1.262801170349121}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:37 - INFO - __main__ - Step 145330: {'lr': 1.2278639215227983e-06, 'samples': 27903360, 'steps': 145329, 'loss/train': 1.262801170349121}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:43 - INFO - __main__ - Step 145341: {'lr': 1.2220923104898364e-06, 'samples': 27905472, 'steps': 145340, 'loss/train': 1.2872779369354248}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:45 - INFO - __main__ - Step 145345: {'lr': 1.2199969056895522e-06, 'samples': 27906240, 'steps': 145344, 'loss/train': 1.8348995447158813}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:45 - INFO - __main__ - Step 145345: {'lr': 1.2199969056895522e-06, 'samples': 27906240, 'steps': 145344, 'loss/train': 1.8348995447158813}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:49 - INFO - __main__ - Step 145353: {'lr': 1.2158114767326922e-06, 'samples': 27907776, 'steps': 145352, 'loss/train': 0.7935240268707275}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:51 - INFO - __main__ - Step 145357: {'lr': 1.2137214526063422e-06, 'samples': 27908544, 'steps': 145356, 'loss/train': 1.474877119064331}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:53 - INFO - __main__ - Step 145362: {'lr': 1.2111114446834437e-06, 'samples': 27909504, 'steps': 145361, 'loss/train': 1.2348850965499878}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:56 - INFO - __main__ - Step 145367: {'lr': 1.2085042392713796e-06, 'samples': 27910464, 'steps': 145366, 'loss/train': 1.477564811706543}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:56 - INFO - __main__ - Step 145367: {'lr': 1.2085042392713796e-06, 'samples': 27910464, 'steps': 145366, 'loss/train': 1.477564811706543}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:41:59 - INFO - __main__ - Step 145374: {'lr': 1.2048588599686828e-06, 'samples': 27911808, 'steps': 145373, 'loss/train': 1.6171029806137085}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:01 - INFO - __main__ - Step 145378: {'lr': 1.2027782523477693e-06, 'samples': 27912576, 'steps': 145377, 'loss/train': 1.02040696144104}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:03 - INFO - __main__ - Step 145382: {'lr': 1.200699438393904e-06, 'samples': 27913344, 'steps': 145381, 'loss/train': 0.9836835265159607}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:06 - INFO - __main__ - Step 145387: {'lr': 1.1981034433186378e-06, 'samples': 27914304, 'steps': 145386, 'loss/train': 1.3758447170257568}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:06 - INFO - __main__ - Step 145387: {'lr': 1.1981034433186378e-06, 'samples': 27914304, 'steps': 145386, 'loss/train': 1.3758447170257568}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:09 - INFO - __main__ - Step 145394: {'lr': 1.1944737586842002e-06, 'samples': 27915648, 'steps': 145393, 'loss/train': 1.0570194721221924}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:11 - INFO - __main__ - Step 145398: {'lr': 1.1924021195481582e-06, 'samples': 27916416, 'steps': 145397, 'loss/train': 0.7255545854568481}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:14 - INFO - __main__ - Step 145403: {'lr': 1.189815093079305e-06, 'samples': 27917376, 'steps': 145402, 'loss/train': 0.7858760952949524}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:16 - INFO - __main__ - Step 145407: {'lr': 1.1877474898833451e-06, 'samples': 27918144, 'steps': 145406, 'loss/train': 1.2728685140609741}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:18 - INFO - __main__ - Step 145411: {'lr': 1.1856816804627912e-06, 'samples': 27918912, 'steps': 145410, 'loss/train': 1.2188702821731567}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:19 - INFO - __main__ - Step 145415: {'lr': 1.1836176648324925e-06, 'samples': 27919680, 'steps': 145414, 'loss/train': 1.094948410987854}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:22 - INFO - __main__ - Step 145419: {'lr': 1.1815554430073538e-06, 'samples': 27920448, 'steps': 145418, 'loss/train': 1.3592917919158936}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:22 - INFO - __main__ - Step 145419: {'lr': 1.1815554430073538e-06, 'samples': 27920448, 'steps': 145418, 'loss/train': 1.3592917919158936}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:25 - INFO - __main__ - Step 145426: {'lr': 1.1779508712017607e-06, 'samples': 27921792, 'steps': 145425, 'loss/train': 1.2662957906723022}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:28 - INFO - __main__ - Step 145431: {'lr': 1.1753795405115009e-06, 'samples': 27922752, 'steps': 145430, 'loss/train': 1.2945735454559326}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:28 - INFO - __main__ - Step 145431: {'lr': 1.1753795405115009e-06, 'samples': 27922752, 'steps': 145430, 'loss/train': 1.2945735454559326}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:32 - INFO - __main__ - Step 145438: {'lr': 1.1717843864461065e-06, 'samples': 27924096, 'steps': 145437, 'loss/train': 0.5848445296287537}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:33 - INFO - __main__ - Step 145442: {'lr': 1.1697324792887544e-06, 'samples': 27924864, 'steps': 145441, 'loss/train': 1.381041169166565}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:35 - INFO - __main__ - Step 145446: {'lr': 1.1676823660366486e-06, 'samples': 27925632, 'steps': 145445, 'loss/train': 1.2882307767868042}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:38 - INFO - __main__ - Step 145451: {'lr': 1.1651222471733281e-06, 'samples': 27926592, 'steps': 145450, 'loss/train': 0.2168867290019989}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:38 - INFO - __main__ - Step 145451: {'lr': 1.1651222471733281e-06, 'samples': 27926592, 'steps': 145450, 'loss/train': 0.2168867290019989}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:41 - INFO - __main__ - Step 145458: {'lr': 1.1615427898596653e-06, 'samples': 27927936, 'steps': 145457, 'loss/train': 1.3662793636322021}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:44 - INFO - __main__ - Step 145462: {'lr': 1.159499852376289e-06, 'samples': 27928704, 'steps': 145461, 'loss/train': 1.2194249629974365}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:46 - INFO - __main__ - Step 145466: {'lr': 1.1574587088719335e-06, 'samples': 27929472, 'steps': 145465, 'loss/train': 0.7195792198181152}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:48 - INFO - __main__ - Step 145470: {'lr': 1.155419359361337e-06, 'samples': 27930240, 'steps': 145469, 'loss/train': 1.831721305847168}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:50 - INFO - __main__ - Step 145474: {'lr': 1.1533818038591826e-06, 'samples': 27931008, 'steps': 145473, 'loss/train': 1.403232216835022}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:52 - INFO - __main__ - Step 145478: {'lr': 1.1513460423801524e-06, 'samples': 27931776, 'steps': 145477, 'loss/train': 1.9069355726242065}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:54 - INFO - __main__ - Step 145483: {'lr': 1.148803863398773e-06, 'samples': 27932736, 'steps': 145482, 'loss/train': 1.3627145290374756}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:54 - INFO - __main__ - Step 145483: {'lr': 1.148803863398773e-06, 'samples': 27932736, 'steps': 145482, 'loss/train': 1.3627145290374756}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:42:54 - INFO - __main__ - Step 145483: {'lr': 1.148803863398773e-06, 'samples': 27932736, 'steps': 145482, 'loss/train': 1.3627145290374756}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:00 - INFO - __main__ - Step 145493: {'lr': 1.1437279151028102e-06, 'samples': 27934656, 'steps': 145492, 'loss/train': 1.3381149768829346}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:02 - INFO - __main__ - Step 145498: {'lr': 1.1411941458453755e-06, 'samples': 27935616, 'steps': 145497, 'loss/train': 1.5607088804244995}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:04 - INFO - __main__ - Step 145502: {'lr': 1.1391691488129296e-06, 'samples': 27936384, 'steps': 145501, 'loss/train': 1.2344887256622314}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:06 - INFO - __main__ - Step 145506: {'lr': 1.1371459459061095e-06, 'samples': 27937152, 'steps': 145505, 'loss/train': 0.9281927347183228}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:06 - INFO - __main__ - Step 145506: {'lr': 1.1371459459061095e-06, 'samples': 27937152, 'steps': 145505, 'loss/train': 0.9281927347183228}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:06 - INFO - __main__ - Step 145506: {'lr': 1.1371459459061095e-06, 'samples': 27937152, 'steps': 145505, 'loss/train': 0.9281927347183228}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:11 - INFO - __main__ - Step 145516: {'lr': 1.1320957880342486e-06, 'samples': 27939072, 'steps': 145515, 'loss/train': 1.4516410827636719}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:14 - INFO - __main__ - Step 145522: {'lr': 1.1290710758263545e-06, 'samples': 27940224, 'steps': 145521, 'loss/train': 1.6191506385803223}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:14 - INFO - __main__ - Step 145522: {'lr': 1.1290710758263545e-06, 'samples': 27940224, 'steps': 145521, 'loss/train': 1.6191506385803223}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:17 - INFO - __main__ - Step 145529: {'lr': 1.1255473471722954e-06, 'samples': 27941568, 'steps': 145528, 'loss/train': 1.476293921470642}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:19 - INFO - __main__ - Step 145533: {'lr': 1.1235362549941453e-06, 'samples': 27942336, 'steps': 145532, 'loss/train': 1.3067481517791748}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:22 - INFO - __main__ - Step 145538: {'lr': 1.1210249129207929e-06, 'samples': 27943296, 'steps': 145537, 'loss/train': 1.5462567806243896}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:22 - INFO - __main__ - Step 145538: {'lr': 1.1210249129207929e-06, 'samples': 27943296, 'steps': 145537, 'loss/train': 1.5462567806243896}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:26 - INFO - __main__ - Step 145546: {'lr': 1.1170125969484912e-06, 'samples': 27944832, 'steps': 145545, 'loss/train': 0.6745740175247192}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:28 - INFO - __main__ - Step 145550: {'lr': 1.115009130383121e-06, 'samples': 27945600, 'steps': 145549, 'loss/train': 1.1050143241882324}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:29 - INFO - __main__ - Step 145554: {'lr': 1.1130074581175431e-06, 'samples': 27946368, 'steps': 145553, 'loss/train': 0.5835559368133545}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:32 - INFO - __main__ - Step 145559: {'lr': 1.1105078910416121e-06, 'samples': 27947328, 'steps': 145558, 'loss/train': 1.805027961730957}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:34 - INFO - __main__ - Step 145563: {'lr': 1.1085102560032534e-06, 'samples': 27948096, 'steps': 145562, 'loss/train': 1.1188064813613892}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:34 - INFO - __main__ - Step 145563: {'lr': 1.1085102560032534e-06, 'samples': 27948096, 'steps': 145562, 'loss/train': 1.1188064813613892}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:37 - INFO - __main__ - Step 145570: {'lr': 1.1050187123413936e-06, 'samples': 27949440, 'steps': 145569, 'loss/train': 1.4884817600250244}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:39 - INFO - __main__ - Step 145574: {'lr': 1.1030260117909086e-06, 'samples': 27950208, 'steps': 145573, 'loss/train': 1.1508110761642456}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:42 - INFO - __main__ - Step 145579: {'lr': 1.1005376594600547e-06, 'samples': 27951168, 'steps': 145578, 'loss/train': 1.2614563703536987}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:44 - INFO - __main__ - Step 145583: {'lr': 1.0985489962986316e-06, 'samples': 27951936, 'steps': 145582, 'loss/train': 1.0993688106536865}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:46 - INFO - __main__ - Step 145587: {'lr': 1.096562127555656e-06, 'samples': 27952704, 'steps': 145586, 'loss/train': 1.4360759258270264}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:46 - INFO - __main__ - Step 145587: {'lr': 1.096562127555656e-06, 'samples': 27952704, 'steps': 145586, 'loss/train': 1.4360759258270264}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:49 - INFO - __main__ - Step 145594: {'lr': 1.0930894251179324e-06, 'samples': 27954048, 'steps': 145593, 'loss/train': 1.2417114973068237}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:49 - INFO - __main__ - Step 145594: {'lr': 1.0930894251179324e-06, 'samples': 27954048, 'steps': 145593, 'loss/train': 1.2417114973068237}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:54 - INFO - __main__ - Step 145603: {'lr': 1.0886325970545786e-06, 'samples': 27955776, 'steps': 145602, 'loss/train': 1.3061085939407349}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:55 - INFO - __main__ - Step 145607: {'lr': 1.0866547006184447e-06, 'samples': 27956544, 'steps': 145606, 'loss/train': 2.016876459121704}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:43:57 - INFO - __main__ - Step 145611: {'lr': 1.0846785986864948e-06, 'samples': 27957312, 'steps': 145610, 'loss/train': 1.6443709135055542}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:00 - INFO - __main__ - Step 145616: {'lr': 1.082210994814814e-06, 'samples': 27958272, 'steps': 145615, 'loss/train': 0.672545850276947}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:00 - INFO - __main__ - Step 145616: {'lr': 1.082210994814814e-06, 'samples': 27958272, 'steps': 145615, 'loss/train': 0.672545850276947}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:04 - INFO - __main__ - Step 145624: {'lr': 1.0782686608744319e-06, 'samples': 27959808, 'steps': 145623, 'loss/train': 1.1929539442062378}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:06 - INFO - __main__ - Step 145628: {'lr': 1.0763001857441446e-06, 'samples': 27960576, 'steps': 145627, 'loss/train': 1.1156774759292603}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:06 - INFO - __main__ - Step 145628: {'lr': 1.0763001857441446e-06, 'samples': 27960576, 'steps': 145627, 'loss/train': 1.1156774759292603}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:10 - INFO - __main__ - Step 145636: {'lr': 1.0723686192342375e-06, 'samples': 27962112, 'steps': 145635, 'loss/train': 1.3533401489257812}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:12 - INFO - __main__ - Step 145640: {'lr': 1.0704055278829838e-06, 'samples': 27962880, 'steps': 145639, 'loss/train': 1.2072798013687134}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:14 - INFO - __main__ - Step 145644: {'lr': 1.0684442311530985e-06, 'samples': 27963648, 'steps': 145643, 'loss/train': 0.9618166089057922}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:16 - INFO - __main__ - Step 145649: {'lr': 1.0659951339485052e-06, 'samples': 27964608, 'steps': 145648, 'loss/train': 1.4415327310562134}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:16 - INFO - __main__ - Step 145649: {'lr': 1.0659951339485052e-06, 'samples': 27964608, 'steps': 145648, 'loss/train': 1.4415327310562134}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:20 - INFO - __main__ - Step 145657: {'lr': 1.0620824110553763e-06, 'samples': 27966144, 'steps': 145656, 'loss/train': 1.129231333732605}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:22 - INFO - __main__ - Step 145661: {'lr': 1.060128741623756e-06, 'samples': 27966912, 'steps': 145660, 'loss/train': 1.601680040359497}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:24 - INFO - __main__ - Step 145665: {'lr': 1.0581768668875836e-06, 'samples': 27967680, 'steps': 145664, 'loss/train': 1.2459033727645874}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:26 - INFO - __main__ - Step 145670: {'lr': 1.0557395472792775e-06, 'samples': 27968640, 'steps': 145669, 'loss/train': 1.2584242820739746}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:29 - INFO - __main__ - Step 145674: {'lr': 1.053791710659252e-06, 'samples': 27969408, 'steps': 145673, 'loss/train': 0.7174002528190613}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:29 - INFO - __main__ - Step 145674: {'lr': 1.053791710659252e-06, 'samples': 27969408, 'steps': 145673, 'loss/train': 0.7174002528190613}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:32 - INFO - __main__ - Step 145681: {'lr': 1.0503873151785081e-06, 'samples': 27970752, 'steps': 145680, 'loss/train': 0.5317294597625732}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:34 - INFO - __main__ - Step 145686: {'lr': 1.047958969302043e-06, 'samples': 27971712, 'steps': 145685, 'loss/train': 1.9132157564163208}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:37 - INFO - __main__ - Step 145691: {'lr': 1.0455334277742756e-06, 'samples': 27972672, 'steps': 145690, 'loss/train': 1.4875950813293457}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:39 - INFO - __main__ - Step 145695: {'lr': 1.0435950137014926e-06, 'samples': 27973440, 'steps': 145694, 'loss/train': 0.9590394496917725}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:41 - INFO - __main__ - Step 145699: {'lr': 1.04165839444334e-06, 'samples': 27974208, 'steps': 145698, 'loss/train': 1.3826751708984375}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:41 - INFO - __main__ - Step 145699: {'lr': 1.04165839444334e-06, 'samples': 27974208, 'steps': 145698, 'loss/train': 1.3826751708984375}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:44 - INFO - __main__ - Step 145706: {'lr': 1.0382736295563023e-06, 'samples': 27975552, 'steps': 145705, 'loss/train': 0.637066125869751}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:47 - INFO - __main__ - Step 145711: {'lr': 1.0358593056962462e-06, 'samples': 27976512, 'steps': 145710, 'loss/train': 1.3404724597930908}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:49 - INFO - __main__ - Step 145715: {'lr': 1.033929865836114e-06, 'samples': 27977280, 'steps': 145714, 'loss/train': 1.3468072414398193}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:50 - INFO - __main__ - Step 145719: {'lr': 1.0320022208602787e-06, 'samples': 27978048, 'steps': 145718, 'loss/train': 1.1402225494384766}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:52 - INFO - __main__ - Step 145723: {'lr': 1.0300763707826455e-06, 'samples': 27978816, 'steps': 145722, 'loss/train': 1.7274466753005981}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:55 - INFO - __main__ - Step 145728: {'lr': 1.027671582282369e-06, 'samples': 27979776, 'steps': 145727, 'loss/train': 1.4001048803329468}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:57 - INFO - __main__ - Step 145732: {'lr': 1.025749770776463e-06, 'samples': 27980544, 'steps': 145731, 'loss/train': 1.331573247909546}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:44:59 - INFO - __main__ - Step 145736: {'lr': 1.023829754213834e-06, 'samples': 27981312, 'steps': 145735, 'loss/train': 0.9296568632125854}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:00 - INFO - __main__ - Step 145740: {'lr': 1.0219115326083327e-06, 'samples': 27982080, 'steps': 145739, 'loss/train': 1.2404841184616089}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:02 - INFO - __main__ - Step 145744: {'lr': 1.0199951059737812e-06, 'samples': 27982848, 'steps': 145743, 'loss/train': 1.1697046756744385}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:02 - INFO - __main__ - Step 145744: {'lr': 1.0199951059737812e-06, 'samples': 27982848, 'steps': 145743, 'loss/train': 1.1697046756744385}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:07 - INFO - __main__ - Step 145752: {'lr': 1.016167637672788e-06, 'samples': 27984384, 'steps': 145751, 'loss/train': 0.9773077964782715}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:08 - INFO - __main__ - Step 145756: {'lr': 1.0142565960339356e-06, 'samples': 27985152, 'steps': 145755, 'loss/train': 1.637543797492981}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:10 - INFO - __main__ - Step 145760: {'lr': 1.0123473494212388e-06, 'samples': 27985920, 'steps': 145759, 'loss/train': 1.18601393699646}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:13 - INFO - __main__ - Step 145765: {'lr': 1.0099633154318499e-06, 'samples': 27986880, 'steps': 145764, 'loss/train': 1.1972029209136963}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:15 - INFO - __main__ - Step 145769: {'lr': 1.0080581076782613e-06, 'samples': 27987648, 'steps': 145768, 'loss/train': 0.69999760389328}3}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:17 - INFO - __main__ - Step 145773: {'lr': 1.0061546949955146e-06, 'samples': 27988416, 'steps': 145772, 'loss/train': 0.8950677514076233}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:18 - INFO - __main__ - Step 145777: {'lr': 1.0042530773973213e-06, 'samples': 27989184, 'steps': 145776, 'loss/train': 1.320884346961975}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:20 - INFO - __main__ - Step 145781: {'lr': 1.0023532548973924e-06, 'samples': 27989952, 'steps': 145780, 'loss/train': 0.9937403798103333}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:23 - INFO - __main__ - Step 145786: {'lr': 9.999810011502575e-07, 'samples': 27990912, 'steps': 145785, 'loss/train': 1.3141206502914429}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:23 - INFO - __main__ - Step 145786: {'lr': 9.999810011502575e-07, 'samples': 27990912, 'steps': 145785, 'loss/train': 1.3141206502914429}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:26 - INFO - __main__ - Step 145793: {'lr': 9.966645581240275e-07, 'samples': 27992256, 'steps': 145792, 'loss/train': 1.3593907356262207}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:28 - INFO - __main__ - Step 145797: {'lr': 9.947719161539637e-07, 'samples': 27993024, 'steps': 145796, 'loss/train': 0.8052687048912048}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:31 - INFO - __main__ - Step 145802: {'lr': 9.924086381460361e-07, 'samples': 27993984, 'steps': 145801, 'loss/train': 1.450958013534546}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:33 - INFO - __main__ - Step 145806: {'lr': 9.905200353200194e-07, 'samples': 27994752, 'steps': 145805, 'loss/train': 1.635334849357605}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:35 - INFO - __main__ - Step 145810: {'lr': 9.886332276912713e-07, 'samples': 27995520, 'steps': 145809, 'loss/train': 1.8607650995254517}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:36 - INFO - __main__ - Step 145814: {'lr': 9.867482152734198e-07, 'samples': 27996288, 'steps': 145813, 'loss/train': 1.1628457307815552}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:38 - INFO - __main__ - Step 145818: {'lr': 9.848649980799817e-07, 'samples': 27997056, 'steps': 145817, 'loss/train': 1.5264832973480225}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:41 - INFO - __main__ - Step 145823: {'lr': 9.82513501143284e-07, 'samples': 27998016, 'steps': 145822, 'loss/train': 1.3766002655029297}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:41 - INFO - __main__ - Step 145823: {'lr': 9.82513501143284e-07, 'samples': 27998016, 'steps': 145822, 'loss/train': 1.3766002655029297}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:45 - INFO - __main__ - Step 145831: {'lr': 9.7875694063429e-07, 'samples': 27999552, 'steps': 145830, 'loss/train': 1.4756295680999756}}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:46 - INFO - __main__ - Step 145835: {'lr': 9.768813532961273e-07, 'samples': 28000320, 'steps': 145834, 'loss/train': 1.1529914140701294}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:48 - INFO - __main__ - Step 145839: {'lr': 9.750075612535714e-07, 'samples': 28001088, 'steps': 145838, 'loss/train': 1.2277498245239258}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:48 - INFO - __main__ - Step 145839: {'lr': 9.750075612535714e-07, 'samples': 28001088, 'steps': 145838, 'loss/train': 1.2277498245239258}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:53 - INFO - __main__ - Step 145847: {'lr': 9.71265363109236e-07, 'samples': 28002624, 'steps': 145846, 'loss/train': 1.2301521301269531}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:55 - INFO - __main__ - Step 145851: {'lr': 9.693969570344629e-07, 'samples': 28003392, 'steps': 145850, 'loss/train': 1.8046475648880005}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:56 - INFO - __main__ - Step 145855: {'lr': 9.675303463092255e-07, 'samples': 28004160, 'steps': 145854, 'loss/train': 1.3237323760986328}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:45:59 - INFO - __main__ - Step 145860: {'lr': 9.651996076334712e-07, 'samples': 28005120, 'steps': 145859, 'loss/train': 0.9237319231033325}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:01 - INFO - __main__ - Step 145865: {'lr': 9.628716742386967e-07, 'samples': 28006080, 'steps': 145864, 'loss/train': 1.055505633354187}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:01 - INFO - __main__ - Step 145865: {'lr': 9.628716742386967e-07, 'samples': 28006080, 'steps': 145864, 'loss/train': 1.055505633354187}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:05 - INFO - __main__ - Step 145873: {'lr': 9.591528158568453e-07, 'samples': 28007616, 'steps': 145872, 'loss/train': 1.038730263710022}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:05 - INFO - __main__ - Step 145873: {'lr': 9.591528158568453e-07, 'samples': 28007616, 'steps': 145872, 'loss/train': 1.038730263710022}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:09 - INFO - __main__ - Step 145880: {'lr': 9.559047060021254e-07, 'samples': 28008960, 'steps': 145879, 'loss/train': 1.0277420282363892}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:09 - INFO - __main__ - Step 145880: {'lr': 9.559047060021254e-07, 'samples': 28008960, 'steps': 145879, 'loss/train': 1.0277420282363892}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:11 - INFO - __main__ - Step 145886: {'lr': 9.531249882398229e-07, 'samples': 28010112, 'steps': 145885, 'loss/train': 0.9693202972412109}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:14 - INFO - __main__ - Step 145892: {'lr': 9.503493102779592e-07, 'samples': 28011264, 'steps': 145891, 'loss/train': 0.9545254111289978}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:16 - INFO - __main__ - Step 145896: {'lr': 9.48501102659205e-07, 'samples': 28012032, 'steps': 145895, 'loss/train': 1.2472002506256104}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:19 - INFO - __main__ - Step 145901: {'lr': 9.461933680594559e-07, 'samples': 28012992, 'steps': 145900, 'loss/train': 1.165945053100586}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:21 - INFO - __main__ - Step 145905: {'lr': 9.443492003348431e-07, 'samples': 28013760, 'steps': 145904, 'loss/train': 1.1532588005065918}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:21 - INFO - __main__ - Step 145905: {'lr': 9.443492003348431e-07, 'samples': 28013760, 'steps': 145904, 'loss/train': 1.1532588005065918}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:24 - INFO - __main__ - Step 145912: {'lr': 9.411262273191501e-07, 'samples': 28015104, 'steps': 145911, 'loss/train': 1.4148972034454346}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:27 - INFO - __main__ - Step 145917: {'lr': 9.388274703949207e-07, 'samples': 28016064, 'steps': 145916, 'loss/train': 1.1223492622375488}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:29 - INFO - __main__ - Step 145922: {'lr': 9.365315190487111e-07, 'samples': 28017024, 'steps': 145921, 'loss/train': 1.2220097780227661}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:29 - INFO - __main__ - Step 145922: {'lr': 9.365315190487111e-07, 'samples': 28017024, 'steps': 145921, 'loss/train': 1.2220097780227661}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:32 - INFO - __main__ - Step 145929: {'lr': 9.333219005843163e-07, 'samples': 28018368, 'steps': 145928, 'loss/train': 1.1463991403579712}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:34 - INFO - __main__ - Step 145933: {'lr': 9.314903018490806e-07, 'samples': 28019136, 'steps': 145932, 'loss/train': 1.3436044454574585}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:37 - INFO - __main__ - Step 145938: {'lr': 9.292033285262236e-07, 'samples': 28020096, 'steps': 145937, 'loss/train': 1.3870861530303955}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:39 - INFO - __main__ - Step 145942: {'lr': 9.273757699609087e-07, 'samples': 28020864, 'steps': 145941, 'loss/train': 0.5918026566505432}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:39 - INFO - __main__ - Step 145942: {'lr': 9.273757699609087e-07, 'samples': 28020864, 'steps': 145941, 'loss/train': 0.5918026566505432}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:42 - INFO - __main__ - Step 145949: {'lr': 9.241818632683419e-07, 'samples': 28022208, 'steps': 145948, 'loss/train': 1.430880069732666}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:45 - INFO - __main__ - Step 145954: {'lr': 9.219038682325986e-07, 'samples': 28023168, 'steps': 145953, 'loss/train': 1.0837693214416504}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:47 - INFO - __main__ - Step 145959: {'lr': 9.196286789655006e-07, 'samples': 28024128, 'steps': 145958, 'loss/train': 1.183017611503601}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:47 - INFO - __main__ - Step 145959: {'lr': 9.196286789655006e-07, 'samples': 28024128, 'steps': 145958, 'loss/train': 1.183017611503601}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:51 - INFO - __main__ - Step 145966: {'lr': 9.164481277317005e-07, 'samples': 28025472, 'steps': 145965, 'loss/train': 1.3857585191726685}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:53 - INFO - __main__ - Step 145970: {'lr': 9.146331390098294e-07, 'samples': 28026240, 'steps': 145969, 'loss/train': 2.2632479667663574}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:55 - INFO - __main__ - Step 145974: {'lr': 9.128199460323172e-07, 'samples': 28027008, 'steps': 145973, 'loss/train': 1.3220419883728027}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:58 - INFO - __main__ - Step 145980: {'lr': 9.101035236153133e-07, 'samples': 28028160, 'steps': 145979, 'loss/train': 0.988226056098938}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:46:58 - INFO - __main__ - Step 145980: {'lr': 9.101035236153133e-07, 'samples': 28028160, 'steps': 145979, 'loss/train': 0.988226056098938}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:02 - INFO - __main__ - Step 145987: {'lr': 9.069394708727807e-07, 'samples': 28029504, 'steps': 145986, 'loss/train': 1.0832916498184204}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:03 - INFO - __main__ - Step 145991: {'lr': 9.051339099544753e-07, 'samples': 28030272, 'steps': 145990, 'loss/train': 0.5870440602302551}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:05 - INFO - __main__ - Step 145995: {'lr': 9.033301448490294e-07, 'samples': 28031040, 'steps': 145994, 'loss/train': 1.3316866159439087}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:05 - INFO - __main__ - Step 145995: {'lr': 9.033301448490294e-07, 'samples': 28031040, 'steps': 145994, 'loss/train': 1.3316866159439087}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:08 - INFO - __main__ - Step 146002: {'lr': 9.001778771282609e-07, 'samples': 28032384, 'steps': 146001, 'loss/train': 1.435117244720459}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:10 - INFO - __main__ - Step 146006: {'lr': 8.983790505751322e-07, 'samples': 28033152, 'steps': 146005, 'loss/train': 1.1360266208648682}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:13 - INFO - __main__ - Step 146011: {'lr': 8.961330428155501e-07, 'samples': 28034112, 'steps': 146010, 'loss/train': 1.037127137184143}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:13 - INFO - __main__ - Step 146011: {'lr': 8.961330428155501e-07, 'samples': 28034112, 'steps': 146010, 'loss/train': 1.037127137184143}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:13 - INFO - __main__ - Step 146011: {'lr': 8.961330428155501e-07, 'samples': 28034112, 'steps': 146010, 'loss/train': 1.037127137184143}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:18 - INFO - __main__ - Step 146022: {'lr': 8.912017031075804e-07, 'samples': 28036224, 'steps': 146021, 'loss/train': 1.3618531227111816}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:21 - INFO - __main__ - Step 146027: {'lr': 8.889646748255764e-07, 'samples': 28037184, 'steps': 146026, 'loss/train': 1.0792341232299805}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:21 - INFO - __main__ - Step 146027: {'lr': 8.889646748255764e-07, 'samples': 28037184, 'steps': 146026, 'loss/train': 1.0792341232299805}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:25 - INFO - __main__ - Step 146035: {'lr': 8.853912663554508e-07, 'samples': 28038720, 'steps': 146034, 'loss/train': 1.1498305797576904}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:27 - INFO - __main__ - Step 146039: {'lr': 8.836072560466923e-07, 'samples': 28039488, 'steps': 146038, 'loss/train': 1.5315619707107544}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:29 - INFO - __main__ - Step 146043: {'lr': 8.818250417060026e-07, 'samples': 28040256, 'steps': 146042, 'loss/train': 1.5979540348052979}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:31 - INFO - __main__ - Step 146047: {'lr': 8.800446233461768e-07, 'samples': 28041024, 'steps': 146046, 'loss/train': 1.8707256317138672}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:31 - INFO - __main__ - Step 146047: {'lr': 8.800446233461768e-07, 'samples': 28041024, 'steps': 146046, 'loss/train': 1.8707256317138672}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:31 - INFO - __main__ - Step 146047: {'lr': 8.800446233461768e-07, 'samples': 28041024, 'steps': 146046, 'loss/train': 1.8707256317138672}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:37 - INFO - __main__ - Step 146058: {'lr': 8.751577334878969e-07, 'samples': 28043136, 'steps': 146057, 'loss/train': 1.563788890838623}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:39 - INFO - __main__ - Step 146063: {'lr': 8.729409099723085e-07, 'samples': 28044096, 'steps': 146062, 'loss/train': 1.0467091798782349}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:42 - INFO - __main__ - Step 146068: {'lr': 8.707268927769518e-07, 'samples': 28045056, 'steps': 146067, 'loss/train': 1.2460633516311646}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:42 - INFO - __main__ - Step 146068: {'lr': 8.707268927769518e-07, 'samples': 28045056, 'steps': 146067, 'loss/train': 1.2460633516311646}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:45 - INFO - __main__ - Step 146074: {'lr': 8.680737765203173e-07, 'samples': 28046208, 'steps': 146073, 'loss/train': 1.5289256572723389}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:47 - INFO - __main__ - Step 146079: {'lr': 8.658659333171392e-07, 'samples': 28047168, 'steps': 146078, 'loss/train': 1.4424527883529663}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:49 - INFO - __main__ - Step 146083: {'lr': 8.641016793613787e-07, 'samples': 28047936, 'steps': 146082, 'loss/train': 1.6112011671066284}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:51 - INFO - __main__ - Step 146087: {'lr': 8.623392215142689e-07, 'samples': 28048704, 'steps': 146086, 'loss/train': 0.9933158159255981}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:53 - INFO - __main__ - Step 146091: {'lr': 8.605785597884941e-07, 'samples': 28049472, 'steps': 146090, 'loss/train': 1.4350316524505615}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:55 - INFO - __main__ - Step 146095: {'lr': 8.588196941968218e-07, 'samples': 28050240, 'steps': 146094, 'loss/train': 1.6032323837280273}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:55 - INFO - __main__ - Step 146095: {'lr': 8.588196941968218e-07, 'samples': 28050240, 'steps': 146094, 'loss/train': 1.6032323837280273}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:47:59 - INFO - __main__ - Step 146102: {'lr': 8.557460013970675e-07, 'samples': 28051584, 'steps': 146101, 'loss/train': 1.3300389051437378}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:01 - INFO - __main__ - Step 146106: {'lr': 8.539920752393959e-07, 'samples': 28052352, 'steps': 146105, 'loss/train': 1.292376160621643}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:03 - INFO - __main__ - Step 146110: {'lr': 8.522399452632613e-07, 'samples': 28053120, 'steps': 146109, 'loss/train': 1.2596211433410645}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:05 - INFO - __main__ - Step 146115: {'lr': 8.500523086926171e-07, 'samples': 28054080, 'steps': 146114, 'loss/train': 1.1976443529129028}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:07 - INFO - __main__ - Step 146119: {'lr': 8.483042201711255e-07, 'samples': 28054848, 'steps': 146118, 'loss/train': 1.4874227046966553}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:07 - INFO - __main__ - Step 146119: {'lr': 8.483042201711255e-07, 'samples': 28054848, 'steps': 146118, 'loss/train': 1.4874227046966553}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:07 - INFO - __main__ - Step 146119: {'lr': 8.483042201711255e-07, 'samples': 28054848, 'steps': 146118, 'loss/train': 1.4874227046966553}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:13 - INFO - __main__ - Step 146128: {'lr': 8.443775884556526e-07, 'samples': 28056576, 'steps': 146127, 'loss/train': 1.3685802221298218}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:15 - INFO - __main__ - Step 146133: {'lr': 8.422000556810583e-07, 'samples': 28057536, 'steps': 146132, 'loss/train': 1.0091112852096558}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:15 - INFO - __main__ - Step 146133: {'lr': 8.422000556810583e-07, 'samples': 28057536, 'steps': 146132, 'loss/train': 1.0091112852096558}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:19 - INFO - __main__ - Step 146140: {'lr': 8.39156225001858e-07, 'samples': 28058880, 'steps': 146139, 'loss/train': 1.3193602561950684}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:21 - INFO - __main__ - Step 146144: {'lr': 8.374193630775461e-07, 'samples': 28059648, 'steps': 146143, 'loss/train': 1.0215989351272583}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:23 - INFO - __main__ - Step 146149: {'lr': 8.352508117219493e-07, 'samples': 28060608, 'steps': 146148, 'loss/train': 1.3044862747192383}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:23 - INFO - __main__ - Step 146149: {'lr': 8.352508117219493e-07, 'samples': 28060608, 'steps': 146148, 'loss/train': 1.3044862747192383}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:27 - INFO - __main__ - Step 146157: {'lr': 8.317869675922574e-07, 'samples': 28062144, 'steps': 146156, 'loss/train': 1.1962308883666992}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:29 - INFO - __main__ - Step 146161: {'lr': 8.300577400336407e-07, 'samples': 28062912, 'steps': 146160, 'loss/train': 0.5279120802879333}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:29 - INFO - __main__ - Step 146161: {'lr': 8.300577400336407e-07, 'samples': 28062912, 'steps': 146160, 'loss/train': 0.5279120802879333}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:33 - INFO - __main__ - Step 146168: {'lr': 8.270359142905637e-07, 'samples': 28064256, 'steps': 146167, 'loss/train': 0.12456241250038147}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:35 - INFO - __main__ - Step 146172: {'lr': 8.253116267356308e-07, 'samples': 28065024, 'steps': 146171, 'loss/train': 1.1127383708953857}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:37 - INFO - __main__ - Step 146176: {'lr': 8.235891355690417e-07, 'samples': 28065792, 'steps': 146175, 'loss/train': 1.344448447227478}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:40 - INFO - __main__ - Step 146181: {'lr': 8.214385478007568e-07, 'samples': 28066752, 'steps': 146180, 'loss/train': 1.245387077331543}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:40 - INFO - __main__ - Step 146181: {'lr': 8.214385478007568e-07, 'samples': 28066752, 'steps': 146180, 'loss/train': 1.245387077331543}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:43 - INFO - __main__ - Step 146188: {'lr': 8.184324405232091e-07, 'samples': 28068096, 'steps': 146187, 'loss/train': 0.6812224984169006}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:45 - INFO - __main__ - Step 146193: {'lr': 8.162885893564786e-07, 'samples': 28069056, 'steps': 146192, 'loss/train': 2.034694194793701}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:45 - INFO - __main__ - Step 146193: {'lr': 8.162885893564786e-07, 'samples': 28069056, 'steps': 146192, 'loss/train': 2.034694194793701}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:49 - INFO - __main__ - Step 146200: {'lr': 8.132919134184736e-07, 'samples': 28070400, 'steps': 146199, 'loss/train': 1.4586222171783447}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:51 - INFO - __main__ - Step 146204: {'lr': 8.11581997317068e-07, 'samples': 28071168, 'steps': 146203, 'loss/train': 0.7662909030914307}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:53 - INFO - __main__ - Step 146208: {'lr': 8.098738777028991e-07, 'samples': 28071936, 'steps': 146207, 'loss/train': 1.3576951026916504}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:55 - INFO - __main__ - Step 146213: {'lr': 8.077412545142437e-07, 'samples': 28072896, 'steps': 146212, 'loss/train': 1.2620291709899902}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:48:58 - INFO - __main__ - Step 146218: {'lr': 8.056114383802605e-07, 'samples': 28073856, 'steps': 146217, 'loss/train': 1.3464670181274414}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:00 - INFO - __main__ - Step 146222: {'lr': 8.039096065685369e-07, 'samples': 28074624, 'steps': 146221, 'loss/train': 1.4819139242172241}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:02 - INFO - __main__ - Step 146226: {'lr': 8.022095712994227e-07, 'samples': 28075392, 'steps': 146225, 'loss/train': 1.1241328716278076}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:03 - INFO - __main__ - Step 146230: {'lr': 8.005113325851576e-07, 'samples': 28076160, 'steps': 146229, 'loss/train': 1.7695392370224}76}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:05 - INFO - __main__ - Step 146234: {'lr': 7.988148904380099e-07, 'samples': 28076928, 'steps': 146233, 'loss/train': 2.6162986755371094}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:08 - INFO - __main__ - Step 146239: {'lr': 7.966968641952011e-07, 'samples': 28077888, 'steps': 146238, 'loss/train': 1.2765089273452759}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:10 - INFO - __main__ - Step 146243: {'lr': 7.950044643687537e-07, 'samples': 28078656, 'steps': 146242, 'loss/train': 1.340530276298523}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:10 - INFO - __main__ - Step 146243: {'lr': 7.950044643687537e-07, 'samples': 28078656, 'steps': 146242, 'loss/train': 1.340530276298523}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:13 - INFO - __main__ - Step 146249: {'lr': 7.924692332706429e-07, 'samples': 28079808, 'steps': 146248, 'loss/train': 1.5764347314834595}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:16 - INFO - __main__ - Step 146255: {'lr': 7.899380445789695e-07, 'samples': 28080960, 'steps': 146254, 'loss/train': 2.498117446899414}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:16 - INFO - __main__ - Step 146255: {'lr': 7.899380445789695e-07, 'samples': 28080960, 'steps': 146254, 'loss/train': 2.498117446899414}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:20 - INFO - __main__ - Step 146262: {'lr': 7.869901003126901e-07, 'samples': 28082304, 'steps': 146261, 'loss/train': 1.3252798318862915}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:22 - INFO - __main__ - Step 146266: {'lr': 7.853080311416016e-07, 'samples': 28083072, 'steps': 146265, 'loss/train': 1.20695960521698}5}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:24 - INFO - __main__ - Step 146270: {'lr': 7.836277586472096e-07, 'samples': 28083840, 'steps': 146269, 'loss/train': 1.4394451379776}}5}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:26 - INFO - __main__ - Step 146275: {'lr': 7.815299446243262e-07, 'samples': 28084800, 'steps': 146274, 'loss/train': 1.3597267866134644}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:28 - INFO - __main__ - Step 146279: {'lr': 7.798537146967988e-07, 'samples': 28085568, 'steps': 146278, 'loss/train': 1.669621467590332}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:30 - INFO - __main__ - Step 146283: {'lr': 7.78179281485325e-07, 'samples': 28086336, 'steps': 146282, 'loss/train': 1.722350001335144}}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:30 - INFO - __main__ - Step 146283: {'lr': 7.78179281485325e-07, 'samples': 28086336, 'steps': 146282, 'loss/train': 1.722350001335144}}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:33 - INFO - __main__ - Step 146290: {'lr': 7.752533467495215e-07, 'samples': 28087680, 'steps': 146289, 'loss/train': 0.9652655124664307}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:33 - INFO - __main__ - Step 146290: {'lr': 7.752533467495215e-07, 'samples': 28087680, 'steps': 146289, 'loss/train': 0.9652655124664307}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:36 - INFO - __main__ - Step 146296: {'lr': 7.727497822639651e-07, 'samples': 28088832, 'steps': 146295, 'loss/train': 1.4715083837509155}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:39 - INFO - __main__ - Step 146302: {'lr': 7.70250260504729e-07, 'samples': 28089984, 'steps': 146301, 'loss/train': 1.3233280181884766}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:41 - INFO - __main__ - Step 146306: {'lr': 7.685861586443354e-07, 'samples': 28090752, 'steps': 146305, 'loss/train': 0.7301026582717896}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:44 - INFO - __main__ - Step 146311: {'lr': 7.6650855806637e-07, 'samples': 28091712, 'steps': 146310, 'loss/train': 1.0935189723968506}6}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:44 - INFO - __main__ - Step 146311: {'lr': 7.6650855806637e-07, 'samples': 28091712, 'steps': 146310, 'loss/train': 1.0935189723968506}6}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:48 - INFO - __main__ - Step 146319: {'lr': 7.63190236791167e-07, 'samples': 28093248, 'steps': 146318, 'loss/train': 1.4506783485412598}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:49 - INFO - __main__ - Step 146323: {'lr': 7.6153377140184e-07, 'samples': 28094016, 'steps': 146322, 'loss/train': 0.8799511194229126}}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:51 - INFO - __main__ - Step 146327: {'lr': 7.598791028606277e-07, 'samples': 28094784, 'steps': 146326, 'loss/train': 1.4106554985046387}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:54 - INFO - __main__ - Step 146332: {'lr': 7.578132940199156e-07, 'samples': 28095744, 'steps': 146331, 'loss/train': 0.5169556140899658}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:56 - INFO - __main__ - Step 146336: {'lr': 7.561626684305378e-07, 'samples': 28096512, 'steps': 146335, 'loss/train': 1.604751706123352}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:49:56 - INFO - __main__ - Step 146336: {'lr': 7.561626684305378e-07, 'samples': 28096512, 'steps': 146335, 'loss/train': 1.604751706123352}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:00 - INFO - __main__ - Step 146343: {'lr': 7.532783974152357e-07, 'samples': 28097856, 'steps': 146342, 'loss/train': 1.5636111497879028}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:02 - INFO - __main__ - Step 146348: {'lr': 7.512215730309524e-07, 'samples': 28098816, 'steps': 146347, 'loss/train': 1.3312050104141235}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:04 - INFO - __main__ - Step 146352: {'lr': 7.495781350601583e-07, 'samples': 28099584, 'steps': 146351, 'loss/train': 1.1762453317642212}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:06 - INFO - __main__ - Step 146356: {'lr': 7.479364940236877e-07, 'samples': 28100352, 'steps': 146355, 'loss/train': 1.0679187774658203}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:08 - INFO - __main__ - Step 146360: {'lr': 7.462966499333368e-07, 'samples': 28101120, 'steps': 146359, 'loss/train': 1.4216258525848389}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:10 - INFO - __main__ - Step 146364: {'lr': 7.446586028009572e-07, 'samples': 28101888, 'steps': 146363, 'loss/train': 1.4915428161621094}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:10 - INFO - __main__ - Step 146364: {'lr': 7.446586028009572e-07, 'samples': 28101888, 'steps': 146363, 'loss/train': 1.4915428161621094}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:14 - INFO - __main__ - Step 146372: {'lr': 7.413878994572964e-07, 'samples': 28103424, 'steps': 146371, 'loss/train': 1.262094497680664}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:16 - INFO - __main__ - Step 146376: {'lr': 7.397552432696076e-07, 'samples': 28104192, 'steps': 146375, 'loss/train': 0.9390880465507507}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:18 - INFO - __main__ - Step 146380: {'lr': 7.381243840870189e-07, 'samples': 28104960, 'steps': 146379, 'loss/train': 1.6285094022750854}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:20 - INFO - __main__ - Step 146385: {'lr': 7.360883371651528e-07, 'samples': 28105920, 'steps': 146384, 'loss/train': 1.37661874294281}4}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:22 - INFO - __main__ - Step 146389: {'lr': 7.344615212870609e-07, 'samples': 28106688, 'steps': 146388, 'loss/train': 0.9647455215454102}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:22 - INFO - __main__ - Step 146389: {'lr': 7.344615212870609e-07, 'samples': 28106688, 'steps': 146388, 'loss/train': 0.9647455215454102}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:25 - INFO - __main__ - Step 146396: {'lr': 7.3161891764284e-07, 'samples': 28108032, 'steps': 146395, 'loss/train': 1.0057172775268555}2}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:28 - INFO - __main__ - Step 146401: {'lr': 7.295918559593051e-07, 'samples': 28108992, 'steps': 146400, 'loss/train': 0.9771812558174133}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:30 - INFO - __main__ - Step 146406: {'lr': 7.275676022107658e-07, 'samples': 28109952, 'steps': 146405, 'loss/train': 1.4322786331176758}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:30 - INFO - __main__ - Step 146406: {'lr': 7.275676022107658e-07, 'samples': 28109952, 'steps': 146405, 'loss/train': 1.4322786331176758}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:34 - INFO - __main__ - Step 146413: {'lr': 7.247383643369798e-07, 'samples': 28111296, 'steps': 146412, 'loss/train': 1.8252004384994507}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:36 - INFO - __main__ - Step 146417: {'lr': 7.231241280074296e-07, 'samples': 28112064, 'steps': 146416, 'loss/train': 1.2696398496627808}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:38 - INFO - __main__ - Step 146421: {'lr': 7.215116888029117e-07, 'samples': 28112832, 'steps': 146420, 'loss/train': 1.3080418109893799}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:41 - INFO - __main__ - Step 146426: {'lr': 7.19498667022006e-07, 'samples': 28113792, 'steps': 146425, 'loss/train': 1.1710014343261719}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:43 - INFO - __main__ - Step 146430: {'lr': 7.178902713912417e-07, 'samples': 28114560, 'steps': 146429, 'loss/train': 1.6107426881790161}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:43 - INFO - __main__ - Step 146430: {'lr': 7.178902713912417e-07, 'samples': 28114560, 'steps': 146429, 'loss/train': 1.6107426881790161}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:46 - INFO - __main__ - Step 146437: {'lr': 7.150799034672528e-07, 'samples': 28115904, 'steps': 146436, 'loss/train': 0.9247349500656128}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:48 - INFO - __main__ - Step 146442: {'lr': 7.130758675217841e-07, 'samples': 28116864, 'steps': 146441, 'loss/train': 0.40443000197410583}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:48 - INFO - __main__ - Step 146442: {'lr': 7.130758675217841e-07, 'samples': 28116864, 'steps': 146441, 'loss/train': 0.40443000197410583}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:53 - INFO - __main__ - Step 146450: {'lr': 7.098752509101803e-07, 'samples': 28118400, 'steps': 146449, 'loss/train': 1.4145394563674927}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:53 - INFO - __main__ - Step 146450: {'lr': 7.098752509101803e-07, 'samples': 28118400, 'steps': 146449, 'loss/train': 1.4145394563674927}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:56 - INFO - __main__ - Step 146457: {'lr': 7.070806085017201e-07, 'samples': 28119744, 'steps': 146456, 'loss/train': 1.258002758026123}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:59 - INFO - __main__ - Step 146462: {'lr': 7.050878051755027e-07, 'samples': 28120704, 'steps': 146461, 'loss/train': 0.3427963852882385}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:50:59 - INFO - __main__ - Step 146462: {'lr': 7.050878051755027e-07, 'samples': 28120704, 'steps': 146461, 'loss/train': 0.3427963852882385}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:02 - INFO - __main__ - Step 146469: {'lr': 7.023025983183273e-07, 'samples': 28122048, 'steps': 146468, 'loss/train': 1.3792251348495483}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:04 - INFO - __main__ - Step 146473: {'lr': 7.007135227921624e-07, 'samples': 28122816, 'steps': 146472, 'loss/train': 1.3156135082244873}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:04 - INFO - __main__ - Step 146473: {'lr': 7.007135227921624e-07, 'samples': 28122816, 'steps': 146472, 'loss/train': 1.3156135082244873}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:08 - INFO - __main__ - Step 146481: {'lr': 6.975407636110043e-07, 'samples': 28124352, 'steps': 146480, 'loss/train': 1.6677881479263306}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:10 - INFO - __main__ - Step 146485: {'lr': 6.959570799789095e-07, 'samples': 28125120, 'steps': 146484, 'loss/train': 1.329735279083252}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:12 - INFO - __main__ - Step 146489: {'lr': 6.943751936676901e-07, 'samples': 28125888, 'steps': 146488, 'loss/train': 0.5325272083282471}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:14 - INFO - __main__ - Step 146494: {'lr': 6.924003632785003e-07, 'samples': 28126848, 'steps': 146493, 'loss/train': 1.4095381498336792}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:16 - INFO - __main__ - Step 146498: {'lr': 6.908225209809049e-07, 'samples': 28127616, 'steps': 146497, 'loss/train': 1.4379372596740723}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:18 - INFO - __main__ - Step 146502: {'lr': 6.892464760412387e-07, 'samples': 28128384, 'steps': 146501, 'loss/train': 0.9494122862815857}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:18 - INFO - __main__ - Step 146502: {'lr': 6.892464760412387e-07, 'samples': 28128384, 'steps': 146501, 'loss/train': 0.9494122862815857}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:22 - INFO - __main__ - Step 146509: {'lr': 6.864927223233785e-07, 'samples': 28129728, 'steps': 146508, 'loss/train': 1.3180333375930786}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:25 - INFO - __main__ - Step 146515: {'lr': 6.841367431274226e-07, 'samples': 28130880, 'steps': 146514, 'loss/train': 1.790961503982544}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:25 - INFO - __main__ - Step 146515: {'lr': 6.841367431274226e-07, 'samples': 28130880, 'steps': 146514, 'loss/train': 1.790961503982544}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:29 - INFO - __main__ - Step 146522: {'lr': 6.813932121083888e-07, 'samples': 28132224, 'steps': 146521, 'loss/train': 1.3052538633346558}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:30 - INFO - __main__ - Step 146526: {'lr': 6.798279515541906e-07, 'samples': 28132992, 'steps': 146525, 'loss/train': 1.0089459419250488}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:32 - INFO - __main__ - Step 146530: {'lr': 6.782644884371636e-07, 'samples': 28133760, 'steps': 146529, 'loss/train': 1.322888970375061}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:35 - INFO - __main__ - Step 146535: {'lr': 6.763126872040559e-07, 'samples': 28134720, 'steps': 146534, 'loss/train': 1.2846029996871948}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:37 - INFO - __main__ - Step 146539: {'lr': 6.74753268361833e-07, 'samples': 28135488, 'steps': 146538, 'loss/train': 1.4243868589401245}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:37 - INFO - __main__ - Step 146539: {'lr': 6.74753268361833e-07, 'samples': 28135488, 'steps': 146538, 'loss/train': 1.4243868589401245}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:40 - INFO - __main__ - Step 146546: {'lr': 6.720286105658391e-07, 'samples': 28136832, 'steps': 146545, 'loss/train': 1.135565996170044}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:42 - INFO - __main__ - Step 146551: {'lr': 6.700857967224527e-07, 'samples': 28137792, 'steps': 146550, 'loss/train': 1.0568063259124756}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:42 - INFO - __main__ - Step 146551: {'lr': 6.700857967224527e-07, 'samples': 28137792, 'steps': 146550, 'loss/train': 1.0568063259124756}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:42 - INFO - __main__ - Step 146551: {'lr': 6.700857967224527e-07, 'samples': 28137792, 'steps': 146550, 'loss/train': 1.0568063259124756}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:48 - INFO - __main__ - Step 146562: {'lr': 6.658214925883977e-07, 'samples': 28139904, 'steps': 146561, 'loss/train': 0.6422065496444702}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:50 - INFO - __main__ - Step 146566: {'lr': 6.642742069114249e-07, 'samples': 28140672, 'steps': 146565, 'loss/train': 1.475317358970642}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:50 - INFO - __main__ - Step 146566: {'lr': 6.642742069114249e-07, 'samples': 28140672, 'steps': 146565, 'loss/train': 1.475317358970642}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:55 - INFO - __main__ - Step 146575: {'lr': 6.607993864448103e-07, 'samples': 28142400, 'steps': 146574, 'loss/train': 1.197298526763916}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:57 - INFO - __main__ - Step 146579: {'lr': 6.59257942843744e-07, 'samples': 28143168, 'steps': 146578, 'loss/train': 1.1780648231506348}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:51:58 - INFO - __main__ - Step 146583: {'lr': 6.57718296828147e-07, 'samples': 28143936, 'steps': 146582, 'loss/train': 5.470465183258057}}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:00 - INFO - __main__ - Step 146587: {'lr': 6.561804484090938e-07, 'samples': 28144704, 'steps': 146586, 'loss/train': 1.1371999979019165}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:00 - INFO - __main__ - Step 146587: {'lr': 6.561804484090938e-07, 'samples': 28144704, 'steps': 146586, 'loss/train': 1.1371999979019165}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:05 - INFO - __main__ - Step 146595: {'lr': 6.531101444049725e-07, 'samples': 28146240, 'steps': 146594, 'loss/train': 1.4575421810150146}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:06 - INFO - __main__ - Step 146599: {'lr': 6.515776888420255e-07, 'samples': 28147008, 'steps': 146598, 'loss/train': 1.5339339971542358}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:08 - INFO - __main__ - Step 146603: {'lr': 6.500470309199202e-07, 'samples': 28147776, 'steps': 146602, 'loss/train': 1.1378040313720703}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:11 - INFO - __main__ - Step 146608: {'lr': 6.481362364665066e-07, 'samples': 28148736, 'steps': 146607, 'loss/train': 1.1409403085708618}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:13 - INFO - __main__ - Step 146613: {'lr': 6.462282508656326e-07, 'samples': 28149696, 'steps': 146612, 'loss/train': 0.9973006844520569}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:15 - INFO - __main__ - Step 146617: {'lr': 6.447038847731957e-07, 'samples': 28150464, 'steps': 146616, 'loss/train': 1.656814455986023}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:16 - INFO - __main__ - Dataset epoch: 2r': 6.447038847731957e-07, 'samples': 28150464, 'steps': 146616, 'loss/train': 1.656814455986023}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:18 - INFO - __main__ - Step 146624: {'lr': 6.420405698105425e-07, 'samples': 28151808, 'steps': 146623, 'loss/train': 1.3685191869735718}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:21 - INFO - __main__ - Step 146629: {'lr': 6.401415726821069e-07, 'samples': 28152768, 'steps': 146628, 'loss/train': 0.540405809879303}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:21 - INFO - __main__ - Step 146629: {'lr': 6.401415726821069e-07, 'samples': 28152768, 'steps': 146628, 'loss/train': 0.540405809879303}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:25 - INFO - __main__ - Step 146636: {'lr': 6.374876957306597e-07, 'samples': 28154112, 'steps': 146635, 'loss/train': 0.8975310325622559}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:25 - INFO - __main__ - Step 146636: {'lr': 6.374876957306597e-07, 'samples': 28154112, 'steps': 146635, 'loss/train': 0.8975310325622559}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:28 - INFO - __main__ - Step 146644: {'lr': 6.344614350378419e-07, 'samples': 28155648, 'steps': 146643, 'loss/train': 1.1121189594268799}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:31 - INFO - __main__ - Step 146650: {'lr': 6.321964586445128e-07, 'samples': 28156800, 'steps': 146649, 'loss/train': 1.2153500318527222}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:31 - INFO - __main__ - Step 146650: {'lr': 6.321964586445128e-07, 'samples': 28156800, 'steps': 146649, 'loss/train': 1.2153500318527222}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:31 - INFO - __main__ - Step 146650: {'lr': 6.321964586445128e-07, 'samples': 28156800, 'steps': 146649, 'loss/train': 1.2153500318527222}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:36 - INFO - __main__ - Step 146660: {'lr': 6.284304869098312e-07, 'samples': 28158720, 'steps': 146659, 'loss/train': 1.302454948425293}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:36 - INFO - __main__ - Step 146660: {'lr': 6.284304869098312e-07, 'samples': 28158720, 'steps': 146659, 'loss/train': 1.302454948425293}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:41 - INFO - __main__ - Step 146668: {'lr': 6.254257996485824e-07, 'samples': 28160256, 'steps': 146667, 'loss/train': 1.230118751525879}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:41 - INFO - __main__ - Step 146668: {'lr': 6.254257996485824e-07, 'samples': 28160256, 'steps': 146667, 'loss/train': 1.230118751525879}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:44 - INFO - __main__ - Step 146675: {'lr': 6.228025974195916e-07, 'samples': 28161600, 'steps': 146674, 'loss/train': 0.6057507395744324}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:46 - INFO - __main__ - Step 146680: {'lr': 6.209322525038697e-07, 'samples': 28162560, 'steps': 146679, 'loss/train': 1.325766682624817}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:46 - INFO - __main__ - Step 146680: {'lr': 6.209322525038697e-07, 'samples': 28162560, 'steps': 146679, 'loss/train': 1.325766682624817}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:46 - INFO - __main__ - Step 146680: {'lr': 6.209322525038697e-07, 'samples': 28162560, 'steps': 146679, 'loss/train': 1.325766682624817}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:52 - INFO - __main__ - Step 146691: {'lr': 6.168273819585901e-07, 'samples': 28164672, 'steps': 146690, 'loss/train': 0.5122979879379272}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:55 - INFO - __main__ - Step 146696: {'lr': 6.149660264256618e-07, 'samples': 28165632, 'steps': 146695, 'loss/train': 0.9136058688163757}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:57 - INFO - __main__ - Step 146700: {'lr': 6.134789646411732e-07, 'samples': 28166400, 'steps': 146699, 'loss/train': 1.5684858560562134}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:52:59 - INFO - __main__ - Step 146704: {'lr': 6.119937007721677e-07, 'samples': 28167168, 'steps': 146703, 'loss/train': 1.4015125036239624}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:00 - INFO - __main__ - Step 146708: {'lr': 6.10510234829359e-07, 'samples': 28167936, 'steps': 146707, 'loss/train': 1.1211116313934326}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:02 - INFO - __main__ - Step 146712: {'lr': 6.090285668234607e-07, 'samples': 28168704, 'steps': 146711, 'loss/train': 0.7869669795036316}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:05 - INFO - __main__ - Step 146717: {'lr': 6.071790101810915e-07, 'samples': 28169664, 'steps': 146716, 'loss/train': 1.4948949813842773}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:05 - INFO - __main__ - Step 146717: {'lr': 6.071790101810915e-07, 'samples': 28169664, 'steps': 146716, 'loss/train': 1.4948949813842773}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:09 - INFO - __main__ - Step 146724: {'lr': 6.045943505338103e-07, 'samples': 28171008, 'steps': 146723, 'loss/train': 1.2141295671463013}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:10 - INFO - __main__ - Step 146728: {'lr': 6.031198743821631e-07, 'samples': 28171776, 'steps': 146727, 'loss/train': 1.0880522727966309}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:13 - INFO - __main__ - Step 146732: {'lr': 6.016471962206893e-07, 'samples': 28172544, 'steps': 146731, 'loss/train': 1.3462024927139282}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:15 - INFO - __main__ - Step 146736: {'lr': 6.001763160599916e-07, 'samples': 28173312, 'steps': 146735, 'loss/train': 1.5259877443313599}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:17 - INFO - __main__ - Step 146740: {'lr': 5.987072339106725e-07, 'samples': 28174080, 'steps': 146739, 'loss/train': 0.8556650280952454}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:18 - INFO - __main__ - Step 146744: {'lr': 5.972399497833625e-07, 'samples': 28174848, 'steps': 146743, 'loss/train': 1.3683645725250244}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:21 - INFO - __main__ - Step 146748: {'lr': 5.957744636886087e-07, 'samples': 28175616, 'steps': 146747, 'loss/train': 1.5737494230270386}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:21 - INFO - __main__ - Step 146748: {'lr': 5.957744636886087e-07, 'samples': 28175616, 'steps': 146747, 'loss/train': 1.5737494230270386}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:25 - INFO - __main__ - Step 146756: {'lr': 5.928488856390136e-07, 'samples': 28177152, 'steps': 146755, 'loss/train': 1.3624240159988403}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:25 - INFO - __main__ - Step 146756: {'lr': 5.928488856390136e-07, 'samples': 28177152, 'steps': 146755, 'loss/train': 1.3624240159988403}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:28 - INFO - __main__ - Step 146763: {'lr': 5.902949047409401e-07, 'samples': 28178496, 'steps': 146762, 'loss/train': 1.3008440732955933}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:31 - INFO - __main__ - Step 146768: {'lr': 5.884740040725533e-07, 'samples': 28179456, 'steps': 146767, 'loss/train': 1.055941104888916}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:33 - INFO - __main__ - Step 146772: {'lr': 5.870193063945706e-07, 'samples': 28180224, 'steps': 146771, 'loss/train': 1.165719747543335}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:35 - INFO - __main__ - Step 146776: {'lr': 5.85566406822835e-07, 'samples': 28180992, 'steps': 146775, 'loss/train': 1.3131957054138184}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:35 - INFO - __main__ - Step 146776: {'lr': 5.85566406822835e-07, 'samples': 28180992, 'steps': 146775, 'loss/train': 1.3131957054138184}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:38 - INFO - __main__ - Step 146783: {'lr': 5.830281592969777e-07, 'samples': 28182336, 'steps': 146782, 'loss/train': 1.6733745336532593}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:41 - INFO - __main__ - Step 146789: {'lr': 5.808569015124943e-07, 'samples': 28183488, 'steps': 146788, 'loss/train': 1.382706642150879}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:43 - INFO - __main__ - Step 146793: {'lr': 5.794116440089869e-07, 'samples': 28184256, 'steps': 146792, 'loss/train': 1.314670205116272}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:45 - INFO - __main__ - Step 146797: {'lr': 5.779681846665996e-07, 'samples': 28185024, 'steps': 146796, 'loss/train': 1.1586246490478516}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:45 - INFO - __main__ - Step 146797: {'lr': 5.779681846665996e-07, 'samples': 28185024, 'steps': 146796, 'loss/train': 1.1586246490478516}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:49 - INFO - __main__ - Step 146804: {'lr': 5.754464576738505e-07, 'samples': 28186368, 'steps': 146803, 'loss/train': 1.0783412456512451}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:51 - INFO - __main__ - Step 146809: {'lr': 5.736485957099913e-07, 'samples': 28187328, 'steps': 146808, 'loss/train': 1.0898240804672241}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:53 - INFO - __main__ - Step 146813: {'lr': 5.722123291158898e-07, 'samples': 28188096, 'steps': 146812, 'loss/train': 1.0581674575805664}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:55 - INFO - __main__ - Step 146817: {'lr': 5.707778607347836e-07, 'samples': 28188864, 'steps': 146816, 'loss/train': 0.910969078540802}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:57 - INFO - __main__ - Step 146821: {'lr': 5.693451905770252e-07, 'samples': 28189632, 'steps': 146820, 'loss/train': 1.2214395999908447}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:53:59 - INFO - __main__ - Step 146825: {'lr': 5.679143186529401e-07, 'samples': 28190400, 'steps': 146824, 'loss/train': 1.14402174949646}7}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:01 - INFO - __main__ - Step 146830: {'lr': 5.661282575296079e-07, 'samples': 28191360, 'steps': 146829, 'loss/train': 1.413188099861145}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:03 - INFO - __main__ - Step 146834: {'lr': 5.647014316689513e-07, 'samples': 28192128, 'steps': 146833, 'loss/train': 1.075175404548645}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:03 - INFO - __main__ - Step 146834: {'lr': 5.647014316689513e-07, 'samples': 28192128, 'steps': 146833, 'loss/train': 1.075175404548645}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:07 - INFO - __main__ - Step 146841: {'lr': 5.622088134993319e-07, 'samples': 28193472, 'steps': 146840, 'loss/train': 1.281372308731079}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:09 - INFO - __main__ - Step 146846: {'lr': 5.604317437309381e-07, 'samples': 28194432, 'steps': 146845, 'loss/train': 1.5668070316314697}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:11 - INFO - __main__ - Step 146851: {'lr': 5.586574838030989e-07, 'samples': 28195392, 'steps': 146850, 'loss/train': 1.425949215888977}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:11 - INFO - __main__ - Step 146851: {'lr': 5.586574838030989e-07, 'samples': 28195392, 'steps': 146850, 'loss/train': 1.425949215888977}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:15 - INFO - __main__ - Step 146858: {'lr': 5.561782404741866e-07, 'samples': 28196736, 'steps': 146857, 'loss/train': 1.2483267784118652}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:17 - INFO - __main__ - Step 146862: {'lr': 5.547640026988266e-07, 'samples': 28197504, 'steps': 146861, 'loss/train': 1.3475432395935059}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:19 - INFO - __main__ - Step 146866: {'lr': 5.533515632622499e-07, 'samples': 28198272, 'steps': 146865, 'loss/train': 1.7337368726730347}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:22 - INFO - __main__ - Step 146871: {'lr': 5.515885428959422e-07, 'samples': 28199232, 'steps': 146870, 'loss/train': 1.4824823141098022}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:22 - INFO - __main__ - Step 146871: {'lr': 5.515885428959422e-07, 'samples': 28199232, 'steps': 146870, 'loss/train': 1.4824823141098022}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:25 - INFO - __main__ - Step 146878: {'lr': 5.491250350870003e-07, 'samples': 28200576, 'steps': 146877, 'loss/train': 0.8560175895690918}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:27 - INFO - __main__ - Step 146882: {'lr': 5.477197891072538e-07, 'samples': 28201344, 'steps': 146881, 'loss/train': 1.515516996383667}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:30 - INFO - __main__ - Step 146887: {'lr': 5.459657606191049e-07, 'samples': 28202304, 'steps': 146886, 'loss/train': 1.2528204917907715}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:32 - INFO - __main__ - Step 146891: {'lr': 5.445645610301364e-07, 'samples': 28203072, 'steps': 146890, 'loss/train': 1.2581528425216675}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:34 - INFO - __main__ - Step 146895: {'lr': 5.431651598534759e-07, 'samples': 28203840, 'steps': 146894, 'loss/train': 1.089673399925232}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:35 - INFO - __main__ - Step 146899: {'lr': 5.417675570992264e-07, 'samples': 28204608, 'steps': 146898, 'loss/train': 1.1480213403701782}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:37 - INFO - __main__ - Step 146903: {'lr': 5.403717527774632e-07, 'samples': 28205376, 'steps': 146902, 'loss/train': 1.5261595249176025}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:40 - INFO - __main__ - Step 146908: {'lr': 5.386295264362407e-07, 'samples': 28206336, 'steps': 146907, 'loss/train': 1.0783356428146362}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:42 - INFO - __main__ - Step 146912: {'lr': 5.372377686243546e-07, 'samples': 28207104, 'steps': 146911, 'loss/train': 1.0940643548965454}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:42 - INFO - __main__ - Step 146912: {'lr': 5.372377686243546e-07, 'samples': 28207104, 'steps': 146911, 'loss/train': 1.0940643548965454}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:45 - INFO - __main__ - Step 146919: {'lr': 5.348065200162511e-07, 'samples': 28208448, 'steps': 146918, 'loss/train': 1.3035752773284912}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:47 - INFO - __main__ - Step 146924: {'lr': 5.330732860195087e-07, 'samples': 28209408, 'steps': 146923, 'loss/train': 0.8049302697181702}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:50 - INFO - __main__ - Step 146929: {'lr': 5.313428621714356e-07, 'samples': 28210368, 'steps': 146928, 'loss/train': 1.5772253274917603}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:50 - INFO - __main__ - Step 146929: {'lr': 5.313428621714356e-07, 'samples': 28210368, 'steps': 146928, 'loss/train': 1.5772253274917603}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:53 - INFO - __main__ - Step 146936: {'lr': 5.289249898710324e-07, 'samples': 28211712, 'steps': 146935, 'loss/train': 1.6031862497329712}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:55 - INFO - __main__ - Step 146940: {'lr': 5.275458215250606e-07, 'samples': 28212480, 'steps': 146939, 'loss/train': 1.2147325277328491}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:55 - INFO - __main__ - Step 146940: {'lr': 5.275458215250606e-07, 'samples': 28212480, 'steps': 146939, 'loss/train': 1.2147325277328491}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:54:59 - INFO - __main__ - Step 146948: {'lr': 5.247928804480994e-07, 'samples': 28214016, 'steps': 146947, 'loss/train': 1.3991676568984985}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:02 - INFO - __main__ - Step 146952: {'lr': 5.234191077369554e-07, 'samples': 28214784, 'steps': 146951, 'loss/train': 1.216329574584961}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:03 - INFO - __main__ - Step 146956: {'lr': 5.22047133590553e-07, 'samples': 28215552, 'steps': 146955, 'loss/train': 1.587267279624939}}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:05 - INFO - __main__ - Step 146960: {'lr': 5.206769580188286e-07, 'samples': 28216320, 'steps': 146959, 'loss/train': 1.8461871147155762}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:08 - INFO - __main__ - Step 146965: {'lr': 5.189667678147936e-07, 'samples': 28217280, 'steps': 146964, 'loss/train': 0.9358727335929871}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:08 - INFO - __main__ - Step 146965: {'lr': 5.189667678147936e-07, 'samples': 28217280, 'steps': 146964, 'loss/train': 0.9358727335929871}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:12 - INFO - __main__ - Step 146973: {'lr': 5.162363089361577e-07, 'samples': 28218816, 'steps': 146972, 'loss/train': 0.978467583656311}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:13 - INFO - __main__ - Step 146977: {'lr': 5.148737774167412e-07, 'samples': 28219584, 'steps': 146976, 'loss/train': 1.3124738931655884}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:15 - INFO - __main__ - Step 146981: {'lr': 5.135130445237113e-07, 'samples': 28220352, 'steps': 146980, 'loss/train': 1.0226856470108032}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:18 - INFO - __main__ - Step 146985: {'lr': 5.121541102668382e-07, 'samples': 28221120, 'steps': 146984, 'loss/train': 1.2934582233428955}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:20 - INFO - __main__ - Step 146989: {'lr': 5.107969746559471e-07, 'samples': 28221888, 'steps': 146988, 'loss/train': 1.3607807159423828}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:20 - INFO - __main__ - Step 146989: {'lr': 5.107969746559471e-07, 'samples': 28221888, 'steps': 146988, 'loss/train': 1.3607807159423828}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:23 - INFO - __main__ - Step 146996: {'lr': 5.084263153581625e-07, 'samples': 28223232, 'steps': 146995, 'loss/train': 1.2601268291473389}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:26 - INFO - __main__ - Step 147002: {'lr': 5.063987059374664e-07, 'samples': 28224384, 'steps': 147001, 'loss/train': 1.3547298908233643}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:26 - INFO - __main__ - Step 147002: {'lr': 5.063987059374664e-07, 'samples': 28224384, 'steps': 147001, 'loss/train': 1.3547298908233643}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:26 - INFO - __main__ - Step 147002: {'lr': 5.063987059374664e-07, 'samples': 28224384, 'steps': 147001, 'loss/train': 1.3547298908233643}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:31 - INFO - __main__ - Step 147013: {'lr': 5.026919331030388e-07, 'samples': 28226496, 'steps': 147012, 'loss/train': 1.220733642578125}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:33 - INFO - __main__ - Step 147017: {'lr': 5.013473882872211e-07, 'samples': 28227264, 'steps': 147016, 'loss/train': 1.5551544427871704}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:36 - INFO - __main__ - Step 147022: {'lr': 4.996692367240096e-07, 'samples': 28228224, 'steps': 147021, 'loss/train': 1.273104190826416}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:38 - INFO - __main__ - Step 147026: {'lr': 4.983287390505353e-07, 'samples': 28228992, 'steps': 147025, 'loss/train': 1.277927279472351}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:40 - INFO - __main__ - Step 147030: {'lr': 4.969900401226857e-07, 'samples': 28229760, 'steps': 147029, 'loss/train': 1.5089244842529297}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:41 - INFO - __main__ - Step 147034: {'lr': 4.956531399501474e-07, 'samples': 28230528, 'steps': 147033, 'loss/train': 0.7999866008758545}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:44 - INFO - __main__ - Step 147038: {'lr': 4.943180385425238e-07, 'samples': 28231296, 'steps': 147037, 'loss/train': 1.2632631063461304}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:46 - INFO - __main__ - Step 147043: {'lr': 4.926516913108803e-07, 'samples': 28232256, 'steps': 147042, 'loss/train': 1.3785791397094727}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:46 - INFO - __main__ - Step 147043: {'lr': 4.926516913108803e-07, 'samples': 28232256, 'steps': 147042, 'loss/train': 1.3785791397094727}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:50 - INFO - __main__ - Step 147050: {'lr': 4.903235270055373e-07, 'samples': 28233600, 'steps': 147049, 'loss/train': 1.27064049243927}7}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:51 - INFO - __main__ - Step 147054: {'lr': 4.889956207538015e-07, 'samples': 28234368, 'steps': 147053, 'loss/train': 1.8046503067016602}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:54 - INFO - __main__ - Step 147059: {'lr': 4.873382675209259e-07, 'samples': 28235328, 'steps': 147058, 'loss/train': 1.477001667022705}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:56 - INFO - __main__ - Step 147063: {'lr': 4.860144086117347e-07, 'samples': 28236096, 'steps': 147062, 'loss/train': 1.37235689163208}}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:56 - INFO - __main__ - Step 147063: {'lr': 4.860144086117347e-07, 'samples': 28236096, 'steps': 147062, 'loss/train': 1.37235689163208}}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:55:59 - INFO - __main__ - Step 147070: {'lr': 4.837019839717127e-07, 'samples': 28237440, 'steps': 147069, 'loss/train': 0.860072672367096}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:01 - INFO - __main__ - Step 147074: {'lr': 4.823830718801159e-07, 'samples': 28238208, 'steps': 147073, 'loss/train': 2.1415069103240967}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:04 - INFO - __main__ - Step 147079: {'lr': 4.807369614144852e-07, 'samples': 28239168, 'steps': 147078, 'loss/train': 1.7807214260101318}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:06 - INFO - __main__ - Step 147084: {'lr': 4.790936616870522e-07, 'samples': 28240128, 'steps': 147083, 'loss/train': 1.3055325746536255}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:06 - INFO - __main__ - Step 147084: {'lr': 4.790936616870522e-07, 'samples': 28240128, 'steps': 147083, 'loss/train': 1.3055325746536255}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:09 - INFO - __main__ - Step 147091: {'lr': 4.7679776414416567e-07, 'samples': 28241472, 'steps': 147090, 'loss/train': 1.0558485984802246}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:11 - INFO - __main__ - Step 147095: {'lr': 4.754882961765239e-07, 'samples': 28242240, 'steps': 147094, 'loss/train': 1.3285629749298096}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:14 - INFO - __main__ - Step 147100: {'lr': 4.738539909356909e-07, 'samples': 28243200, 'steps': 147099, 'loss/train': 1.6712294816970825}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:14 - INFO - __main__ - Step 147100: {'lr': 4.738539909356909e-07, 'samples': 28243200, 'steps': 147099, 'loss/train': 1.6712294816970825}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:18 - INFO - __main__ - Step 147108: {'lr': 4.7124494905484425e-07, 'samples': 28244736, 'steps': 147107, 'loss/train': 1.3176201581954956}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:20 - INFO - __main__ - Step 147112: {'lr': 4.6994312652107185e-07, 'samples': 28245504, 'steps': 147111, 'loss/train': 1.2434377670288086}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:21 - INFO - __main__ - Step 147116: {'lr': 4.686431029375937e-07, 'samples': 28246272, 'steps': 147115, 'loss/train': 0.9388652443885803}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:24 - INFO - __main__ - Step 147120: {'lr': 4.6734487831376346e-07, 'samples': 28247040, 'steps': 147119, 'loss/train': 1.2424429655075073}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:26 - INFO - __main__ - Step 147125: {'lr': 4.657246273352678e-07, 'samples': 28248000, 'steps': 147124, 'loss/train': 1.2569608688354492}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:28 - INFO - __main__ - Step 147129: {'lr': 4.6443045040489575e-07, 'samples': 28248768, 'steps': 147128, 'loss/train': 1.1841111183166504}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:28 - INFO - __main__ - Step 147129: {'lr': 4.6443045040489575e-07, 'samples': 28248768, 'steps': 147128, 'loss/train': 1.1841111183166504}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:32 - INFO - __main__ - Step 147136: {'lr': 4.6216996960210533e-07, 'samples': 28250112, 'steps': 147135, 'loss/train': 1.6424862146377563}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:34 - INFO - __main__ - Step 147141: {'lr': 4.6055871359118594e-07, 'samples': 28251072, 'steps': 147140, 'loss/train': 1.4060261249542236}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:36 - INFO - __main__ - Step 147145: {'lr': 4.5927173267679144e-07, 'samples': 28251840, 'steps': 147144, 'loss/train': 1.5399407148361206}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:38 - INFO - __main__ - Step 147149: {'lr': 4.5798655078960196e-07, 'samples': 28252608, 'steps': 147148, 'loss/train': 1.3755792379379272}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:40 - INFO - __main__ - Step 147153: {'lr': 4.5670316793891554e-07, 'samples': 28253376, 'steps': 147152, 'loss/train': 1.4204877614974976}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:42 - INFO - __main__ - Step 147157: {'lr': 4.5542158413397484e-07, 'samples': 28254144, 'steps': 147156, 'loss/train': 1.489708423614502}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:44 - INFO - __main__ - Step 147162: {'lr': 4.538221342999249e-07, 'samples': 28255104, 'steps': 147161, 'loss/train': 1.1528587341308594}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:44 - INFO - __main__ - Step 147162: {'lr': 4.538221342999249e-07, 'samples': 28255104, 'steps': 147161, 'loss/train': 1.1528587341308594}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:48 - INFO - __main__ - Step 147169: {'lr': 4.515876270859143e-07, 'samples': 28256448, 'steps': 147168, 'loss/train': 1.0762087106704712}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:50 - INFO - __main__ - Step 147173: {'lr': 4.503132395561882e-07, 'samples': 28257216, 'steps': 147172, 'loss/train': 1.5721454620361328}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:52 - INFO - __main__ - Step 147178: {'lr': 4.487227851179554e-07, 'samples': 28258176, 'steps': 147177, 'loss/train': 1.3311965465545654}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:55 - INFO - __main__ - Step 147183: {'lr': 4.4713514177860404e-07, 'samples': 28259136, 'steps': 147182, 'loss/train': 1.249375343322754}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:55 - INFO - __main__ - Step 147183: {'lr': 4.4713514177860404e-07, 'samples': 28259136, 'steps': 147182, 'loss/train': 1.249375343322754}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:56:58 - INFO - __main__ - Step 147190: {'lr': 4.4491716378364245e-07, 'samples': 28260480, 'steps': 147189, 'loss/train': 1.157741904258728}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:00 - INFO - __main__ - Step 147194: {'lr': 4.4365222158812556e-07, 'samples': 28261248, 'steps': 147193, 'loss/train': 1.3581422567367554}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:03 - INFO - __main__ - Step 147199: {'lr': 4.4207357388514893e-07, 'samples': 28262208, 'steps': 147198, 'loss/train': 1.1158151626586914}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:03 - INFO - __main__ - Step 147199: {'lr': 4.4207357388514893e-07, 'samples': 28262208, 'steps': 147198, 'loss/train': 1.1158151626586914}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:07 - INFO - __main__ - Step 147206: {'lr': 4.398681898769019e-07, 'samples': 28263552, 'steps': 147205, 'loss/train': 0.6801626682281494}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:08 - INFO - __main__ - Step 147210: {'lr': 4.386104442952732e-07, 'samples': 28264320, 'steps': 147209, 'loss/train': 1.3019236326217651}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:10 - INFO - __main__ - Step 147214: {'lr': 4.373544978898414e-07, 'samples': 28265088, 'steps': 147213, 'loss/train': 1.2783881425857544}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:13 - INFO - __main__ - Step 147219: {'lr': 4.357870949883158e-07, 'samples': 28266048, 'steps': 147218, 'loss/train': 1.3335086107254028}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:13 - INFO - __main__ - Step 147219: {'lr': 4.357870949883158e-07, 'samples': 28266048, 'steps': 147218, 'loss/train': 1.3335086107254028}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:16 - INFO - __main__ - Step 147226: {'lr': 4.335974538210441e-07, 'samples': 28267392, 'steps': 147225, 'loss/train': 1.8125379085540771}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:18 - INFO - __main__ - Step 147230: {'lr': 4.323487042107166e-07, 'samples': 28268160, 'steps': 147229, 'loss/train': 1.1549115180969238}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:21 - INFO - __main__ - Step 147235: {'lr': 4.307902973538502e-07, 'samples': 28269120, 'steps': 147234, 'loss/train': 1.10246741771698}8}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:23 - INFO - __main__ - Step 147239: {'lr': 4.2954559600413833e-07, 'samples': 28269888, 'steps': 147238, 'loss/train': 1.272863745689392}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:23 - INFO - __main__ - Step 147239: {'lr': 4.2954559600413833e-07, 'samples': 28269888, 'steps': 147238, 'loss/train': 1.272863745689392}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:26 - INFO - __main__ - Step 147246: {'lr': 4.2737169807244293e-07, 'samples': 28271232, 'steps': 147245, 'loss/train': 0.8144082427024841}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:28 - INFO - __main__ - Step 147251: {'lr': 4.258222874400086e-07, 'samples': 28272192, 'steps': 147250, 'loss/train': 3.0743892192840576}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:28 - INFO - __main__ - Step 147251: {'lr': 4.258222874400086e-07, 'samples': 28272192, 'steps': 147250, 'loss/train': 3.0743892192840576}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:28 - INFO - __main__ - Step 147251: {'lr': 4.258222874400086e-07, 'samples': 28272192, 'steps': 147250, 'loss/train': 3.0743892192840576}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:34 - INFO - __main__ - Step 147262: {'lr': 4.224234800491722e-07, 'samples': 28274304, 'steps': 147261, 'loss/train': 1.386408805847168}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:36 - INFO - __main__ - Step 147267: {'lr': 4.2088306581986034e-07, 'samples': 28275264, 'steps': 147266, 'loss/train': 0.7447375059127808}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:39 - INFO - __main__ - Step 147272: {'lr': 4.193454630030125e-07, 'samples': 28276224, 'steps': 147271, 'loss/train': 0.9925128817558289}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:41 - INFO - __main__ - Step 147277: {'lr': 4.1781067161594797e-07, 'samples': 28277184, 'steps': 147276, 'loss/train': 1.2510809898376465}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:41 - INFO - __main__ - Step 147277: {'lr': 4.1781067161594797e-07, 'samples': 28277184, 'steps': 147276, 'loss/train': 1.2510809898376465}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:41 - INFO - __main__ - Step 147277: {'lr': 4.1781067161594797e-07, 'samples': 28277184, 'steps': 147276, 'loss/train': 1.2510809898376465}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:46 - INFO - __main__ - Step 147287: {'lr': 4.147495232003362e-07, 'samples': 28279104, 'steps': 147286, 'loss/train': 0.7900604009628296}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:48 - INFO - __main__ - Step 147292: {'lr': 4.1322316620623356e-07, 'samples': 28280064, 'steps': 147291, 'loss/train': 1.3049287796020508}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:51 - INFO - __main__ - Step 147296: {'lr': 4.1200410488925753e-07, 'samples': 28280832, 'steps': 147295, 'loss/train': 1.334334135055542}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:53 - INFO - __main__ - Step 147300: {'lr': 4.107868429402417e-07, 'samples': 28281600, 'steps': 147299, 'loss/train': 0.885260820388794}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:55 - INFO - __main__ - Step 147304: {'lr': 4.095713803680123e-07, 'samples': 28282368, 'steps': 147303, 'loss/train': 1.2701653242111206}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:56 - INFO - __main__ - Step 147308: {'lr': 4.083577171813402e-07, 'samples': 28283136, 'steps': 147307, 'loss/train': 1.5522445440292358}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:58 - INFO - __main__ - Step 147312: {'lr': 4.071458533889127e-07, 'samples': 28283904, 'steps': 147311, 'loss/train': 0.7178853750228882}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:58 - INFO - __main__ - Step 147312: {'lr': 4.071458533889127e-07, 'samples': 28283904, 'steps': 147311, 'loss/train': 0.7178853750228882}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:57:58 - INFO - __main__ - Step 147312: {'lr': 4.071458533889127e-07, 'samples': 28283904, 'steps': 147311, 'loss/train': 0.7178853750228882}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:04 - INFO - __main__ - Step 147323: {'lr': 4.0382250615780823e-07, 'samples': 28286016, 'steps': 147322, 'loss/train': 1.1089190244674683}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:04 - INFO - __main__ - Step 147323: {'lr': 4.0382250615780823e-07, 'samples': 28286016, 'steps': 147322, 'loss/train': 1.1089190244674683}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:08 - INFO - __main__ - Step 147331: {'lr': 4.014140736211258e-07, 'samples': 28287552, 'steps': 147330, 'loss/train': 1.3483388423919678}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:10 - INFO - __main__ - Step 147335: {'lr': 4.0021255651498387e-07, 'samples': 28288320, 'steps': 147334, 'loss/train': 1.0245035886764526}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:13 - INFO - __main__ - Step 147340: {'lr': 3.987131906140751e-07, 'samples': 28289280, 'steps': 147339, 'loss/train': 1.1962558031082153}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:13 - INFO - __main__ - Step 147340: {'lr': 3.987131906140751e-07, 'samples': 28289280, 'steps': 147339, 'loss/train': 1.1962558031082153}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:17 - INFO - __main__ - Step 147347: {'lr': 3.966188019491057e-07, 'samples': 28290624, 'steps': 147346, 'loss/train': 0.8226308226585388}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:18 - INFO - __main__ - Step 147351: {'lr': 3.954244827067965e-07, 'samples': 28291392, 'steps': 147350, 'loss/train': 1.4518084526062012}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:20 - INFO - __main__ - Step 147355: {'lr': 3.942319629520186e-07, 'samples': 28292160, 'steps': 147354, 'loss/train': 1.5849599838256836}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:20 - INFO - __main__ - Step 147355: {'lr': 3.942319629520186e-07, 'samples': 28292160, 'steps': 147354, 'loss/train': 1.5849599838256836}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:25 - INFO - __main__ - Step 147363: {'lr': 3.9185232193936214e-07, 'samples': 28293696, 'steps': 147362, 'loss/train': 1.0411624908447266}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:26 - INFO - __main__ - Step 147367: {'lr': 3.9066520069869217e-07, 'samples': 28294464, 'steps': 147366, 'loss/train': 1.1404656171798706}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:28 - INFO - __main__ - Step 147371: {'lr': 3.8947987897980377e-07, 'samples': 28295232, 'steps': 147370, 'loss/train': 1.0294963121414185}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:31 - INFO - __main__ - Step 147376: {'lr': 3.8800075742184695e-07, 'samples': 28296192, 'steps': 147375, 'loss/train': 1.0333302021026611}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:33 - INFO - __main__ - Step 147380: {'lr': 3.8681948465832396e-07, 'samples': 28296960, 'steps': 147379, 'loss/train': 1.3923137187957764}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:35 - INFO - __main__ - Step 147384: {'lr': 3.856400114443659e-07, 'samples': 28297728, 'steps': 147383, 'loss/train': 1.0863401889801025}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:35 - INFO - __main__ - Step 147384: {'lr': 3.856400114443659e-07, 'samples': 28297728, 'steps': 147383, 'loss/train': 1.0863401889801025}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:38 - INFO - __main__ - Step 147391: {'lr': 3.835802635116237e-07, 'samples': 28299072, 'steps': 147390, 'loss/train': 1.1838154792785645}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:41 - INFO - __main__ - Step 147397: {'lr': 3.8181915174079204e-07, 'samples': 28300224, 'steps': 147396, 'loss/train': 0.8638104796409607}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:43 - INFO - __main__ - Step 147401: {'lr': 3.8064732670717304e-07, 'samples': 28300992, 'steps': 147400, 'loss/train': 1.4230722188949585}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:45 - INFO - __main__ - Step 147405: {'lr': 3.794773012675834e-07, 'samples': 28301760, 'steps': 147404, 'loss/train': 1.4542431831359863}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:45 - INFO - __main__ - Step 147405: {'lr': 3.794773012675834e-07, 'samples': 28301760, 'steps': 147404, 'loss/train': 1.4542431831359863}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:48 - INFO - __main__ - Step 147412: {'lr': 3.7743408704687953e-07, 'samples': 28303104, 'steps': 147411, 'loss/train': 1.3053563833236694}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:51 - INFO - __main__ - Step 147417: {'lr': 3.7597802259745075e-07, 'samples': 28304064, 'steps': 147416, 'loss/train': 1.3845674991607666}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:53 - INFO - __main__ - Step 147422: {'lr': 3.7452477006633124e-07, 'samples': 28305024, 'steps': 147421, 'loss/train': 1.3713163137435913}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:53 - INFO - __main__ - Step 147422: {'lr': 3.7452477006633124e-07, 'samples': 28305024, 'steps': 147421, 'loss/train': 1.3713163137435913}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:57 - INFO - __main__ - Step 147428: {'lr': 3.7278457878422987e-07, 'samples': 28306176, 'steps': 147427, 'loss/train': 1.7092797756195068}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:58:57 - INFO - __main__ - Step 147428: {'lr': 3.7278457878422987e-07, 'samples': 28306176, 'steps': 147427, 'loss/train': 1.7092797756195068}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:01 - INFO - __main__ - Step 147436: {'lr': 3.704706225238497e-07, 'samples': 28307712, 'steps': 147435, 'loss/train': 1.2722135782241821}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:03 - INFO - __main__ - Step 147440: {'lr': 3.693163438906189e-07, 'samples': 28308480, 'steps': 147439, 'loss/train': 1.372098684310913}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:03 - INFO - __main__ - Step 147440: {'lr': 3.693163438906189e-07, 'samples': 28308480, 'steps': 147439, 'loss/train': 1.372098684310913}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:07 - INFO - __main__ - Step 147448: {'lr': 3.670131856597092e-07, 'samples': 28310016, 'steps': 147447, 'loss/train': 1.0850006341934204}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:07 - INFO - __main__ - Step 147448: {'lr': 3.670131856597092e-07, 'samples': 28310016, 'steps': 147447, 'loss/train': 1.0850006341934204}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:10 - INFO - __main__ - Step 147455: {'lr': 3.6500382744589224e-07, 'samples': 28311360, 'steps': 147454, 'loss/train': 1.574283480644226}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:13 - INFO - __main__ - Step 147460: {'lr': 3.6357194602662514e-07, 'samples': 28312320, 'steps': 147459, 'loss/train': 0.217850923538208}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:13 - INFO - __main__ - Step 147460: {'lr': 3.6357194602662514e-07, 'samples': 28312320, 'steps': 147459, 'loss/train': 0.217850923538208}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:16 - INFO - __main__ - Step 147467: {'lr': 3.615720363007724e-07, 'samples': 28313664, 'steps': 147466, 'loss/train': 1.1487256288528442}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:18 - INFO - __main__ - Step 147471: {'lr': 3.6043170537122807e-07, 'samples': 28314432, 'steps': 147470, 'loss/train': 1.9718769788742065}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:21 - INFO - __main__ - Step 147476: {'lr': 3.5900882259434777e-07, 'samples': 28315392, 'steps': 147475, 'loss/train': 1.262515902519226}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:23 - INFO - __main__ - Step 147480: {'lr': 3.578725410909034e-07, 'samples': 28316160, 'steps': 147479, 'loss/train': 1.4772089719772339}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:25 - INFO - __main__ - Step 147484: {'lr': 3.5673805934569035e-07, 'samples': 28316928, 'steps': 147483, 'loss/train': 2.1232197284698486}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:27 - INFO - __main__ - Step 147488: {'lr': 3.5560537736689656e-07, 'samples': 28317696, 'steps': 147487, 'loss/train': 1.3047112226486206}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:28 - INFO - __main__ - Step 147492: {'lr': 3.544744951626822e-07, 'samples': 28318464, 'steps': 147491, 'loss/train': 1.0067059993743896}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:31 - INFO - __main__ - Step 147496: {'lr': 3.53345412741235e-07, 'samples': 28319232, 'steps': 147495, 'loss/train': 1.422754168510437}6}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:33 - INFO - __main__ - Step 147501: {'lr': 3.5193659067131456e-07, 'samples': 28320192, 'steps': 147500, 'loss/train': 1.2804608345031738}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:35 - INFO - __main__ - Step 147506: {'lr': 3.505305807904868e-07, 'samples': 28321152, 'steps': 147505, 'loss/train': 1.0613142251968384}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:35 - INFO - __main__ - Step 147506: {'lr': 3.505305807904868e-07, 'samples': 28321152, 'steps': 147505, 'loss/train': 1.0613142251968384}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:38 - INFO - __main__ - Step 147512: {'lr': 3.4884708104546137e-07, 'samples': 28322304, 'steps': 147511, 'loss/train': 1.0240473747253418}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:41 - INFO - __main__ - Step 147517: {'lr': 3.474472580364263e-07, 'samples': 28323264, 'steps': 147516, 'loss/train': 1.4320582151412964}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:43 - INFO - __main__ - Step 147521: {'lr': 3.4632942444101647e-07, 'samples': 28324032, 'steps': 147520, 'loss/train': 1.0351128578186035}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:45 - INFO - __main__ - Step 147525: {'lr': 3.4521339068707694e-07, 'samples': 28324800, 'steps': 147524, 'loss/train': 0.5914286375045776}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:47 - INFO - __main__ - Step 147529: {'lr': 3.440991567826568e-07, 'samples': 28325568, 'steps': 147528, 'loss/train': 0.9515489339828491}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:48 - INFO - __main__ - Step 147533: {'lr': 3.4298672273577744e-07, 'samples': 28326336, 'steps': 147532, 'loss/train': 1.4323700666427612}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:51 - INFO - __main__ - Step 147537: {'lr': 3.4187608855443255e-07, 'samples': 28327104, 'steps': 147536, 'loss/train': 1.47052001953125}2}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:53 - INFO - __main__ - Step 147542: {'lr': 3.4049032690092184e-07, 'samples': 28328064, 'steps': 147541, 'loss/train': 1.184960961341858}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:53 - INFO - __main__ - Step 147542: {'lr': 3.4049032690092184e-07, 'samples': 28328064, 'steps': 147541, 'loss/train': 1.184960961341858}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:56 - INFO - __main__ - Step 147549: {'lr': 3.3855498528384545e-07, 'samples': 28329408, 'steps': 147548, 'loss/train': 1.3956201076507568}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 17:59:58 - INFO - __main__ - Step 147553: {'lr': 3.374515506447129e-07, 'samples': 28330176, 'steps': 147552, 'loss/train': 0.553237795829773}8}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:01 - INFO - __main__ - Step 147558: {'lr': 3.3607478846381377e-07, 'samples': 28331136, 'steps': 147557, 'loss/train': 1.2077330350875854}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:01 - INFO - __main__ - Step 147558: {'lr': 3.3607478846381377e-07, 'samples': 28331136, 'steps': 147557, 'loss/train': 1.2077330350875854}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:05 - INFO - __main__ - Step 147566: {'lr': 3.338778187058078e-07, 'samples': 28332672, 'steps': 147565, 'loss/train': 1.142945647239685}4}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:06 - INFO - __main__ - Step 147570: {'lr': 3.327820337196974e-07, 'samples': 28333440, 'steps': 147569, 'loss/train': 0.9832907319068909}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:09 - INFO - __main__ - Step 147574: {'lr': 3.3168804867270143e-07, 'samples': 28334208, 'steps': 147573, 'loss/train': 1.3128719329833984}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:11 - INFO - __main__ - Step 147579: {'lr': 3.3032309854039153e-07, 'samples': 28335168, 'steps': 147578, 'loss/train': 1.1697593927383423}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:13 - INFO - __main__ - Step 147583: {'lr': 3.2923316338528964e-07, 'samples': 28335936, 'steps': 147582, 'loss/train': 1.2015302181243896}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:13 - INFO - __main__ - Step 147583: {'lr': 3.2923316338528964e-07, 'samples': 28335936, 'steps': 147582, 'loss/train': 1.2015302181243896}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:17 - INFO - __main__ - Step 147590: {'lr': 3.2733010803365217e-07, 'samples': 28337280, 'steps': 147589, 'loss/train': 1.2991634607315063}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:19 - INFO - __main__ - Step 147595: {'lr': 3.259741577396058e-07, 'samples': 28338240, 'steps': 147594, 'loss/train': 1.1775381565093994}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:21 - INFO - __main__ - Step 147600: {'lr': 3.2462101992714045e-07, 'samples': 28339200, 'steps': 147599, 'loss/train': 1.2935291528701782}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:21 - INFO - __main__ - Step 147600: {'lr': 3.2462101992714045e-07, 'samples': 28339200, 'steps': 147599, 'loss/train': 1.2935291528701782}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:25 - INFO - __main__ - Step 147607: {'lr': 3.2273135198776505e-07, 'samples': 28340544, 'steps': 147606, 'loss/train': 1.2735909223556519}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:26 - INFO - __main__ - Step 147611: {'lr': 3.216540167499671e-07, 'samples': 28341312, 'steps': 147610, 'loss/train': 1.222668170928955}9}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:29 - INFO - __main__ - Step 147615: {'lr': 3.205784815315249e-07, 'samples': 28342080, 'steps': 147614, 'loss/train': 1.2364749908447266}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:31 - INFO - __main__ - Step 147620: {'lr': 3.192365937974839e-07, 'samples': 28343040, 'steps': 147619, 'loss/train': 1.0921530723571777}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:31 - INFO - __main__ - Step 147620: {'lr': 3.192365937974839e-07, 'samples': 28343040, 'steps': 147619, 'loss/train': 1.0921530723571777}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:31 - INFO - __main__ - Step 147620: {'lr': 3.192365937974839e-07, 'samples': 28343040, 'steps': 147619, 'loss/train': 1.0921530723571777}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:37 - INFO - __main__ - Step 147631: {'lr': 3.162943410061625e-07, 'samples': 28345152, 'steps': 147630, 'loss/train': 1.328928828239441}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:39 - INFO - __main__ - Step 147636: {'lr': 3.1496145350912163e-07, 'samples': 28346112, 'steps': 147635, 'loss/train': 1.3446049690246582}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:42 - INFO - __main__ - Step 147641: {'lr': 3.1363137861770143e-07, 'samples': 28347072, 'steps': 147640, 'loss/train': 0.8127176761627197}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:44 - INFO - __main__ - Step 147645: {'lr': 3.1256934379067513e-07, 'samples': 28347840, 'steps': 147644, 'loss/train': 1.1714727878570557}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:44 - INFO - __main__ - Step 147645: {'lr': 3.1256934379067513e-07, 'samples': 28347840, 'steps': 147644, 'loss/train': 1.1714727878570557}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:47 - INFO - __main__ - Step 147652: {'lr': 3.107151143021958e-07, 'samples': 28349184, 'steps': 147651, 'loss/train': 1.4520893096923828}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:49 - INFO - __main__ - Step 147657: {'lr': 3.093940398493367e-07, 'samples': 28350144, 'steps': 147656, 'loss/train': 0.9486633539199829}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:52 - INFO - __main__ - Step 147662: {'lr': 3.080757780648258e-07, 'samples': 28351104, 'steps': 147661, 'loss/train': 1.2911559343338013}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:52 - INFO - __main__ - Step 147662: {'lr': 3.080757780648258e-07, 'samples': 28351104, 'steps': 147661, 'loss/train': 1.2911559343338013}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:55 - INFO - __main__ - Step 147669: {'lr': 3.062349368776196e-07, 'samples': 28352448, 'steps': 147668, 'loss/train': 1.0389649868011475}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:57 - INFO - __main__ - Step 147673: {'lr': 3.0518550280461686e-07, 'samples': 28353216, 'steps': 147672, 'loss/train': 1.3760560750961304}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:00:59 - INFO - __main__ - Step 147678: {'lr': 3.038762416584795e-07, 'samples': 28354176, 'steps': 147677, 'loss/train': 0.925425112247467}4}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:02 - INFO - __main__ - Step 147682: {'lr': 3.028308579068051e-07, 'samples': 28354944, 'steps': 147681, 'loss/train': 1.8284255266189575}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:04 - INFO - __main__ - Step 147686: {'lr': 3.017872743101835e-07, 'samples': 28355712, 'steps': 147685, 'loss/train': 0.9723736643791199}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:06 - INFO - __main__ - Step 147690: {'lr': 3.007454908761642e-07, 'samples': 28356480, 'steps': 147689, 'loss/train': 1.406026840209961}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:07 - INFO - __main__ - Step 147694: {'lr': 2.997055076122412e-07, 'samples': 28357248, 'steps': 147693, 'loss/train': 1.3686281442642212}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:10 - INFO - __main__ - Step 147699: {'lr': 2.9840806003295084e-07, 'samples': 28358208, 'steps': 147698, 'loss/train': 1.2908576726913452}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:12 - INFO - __main__ - Step 147703: {'lr': 2.9737212717911857e-07, 'samples': 28358976, 'steps': 147702, 'loss/train': 1.116593837738037}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:12 - INFO - __main__ - Step 147703: {'lr': 2.9737212717911857e-07, 'samples': 28358976, 'steps': 147702, 'loss/train': 1.116593837738037}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:15 - INFO - __main__ - Step 147710: {'lr': 2.955635764072262e-07, 'samples': 28360320, 'steps': 147709, 'loss/train': 1.3437886238098145}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:17 - INFO - __main__ - Step 147714: {'lr': 2.945325941059729e-07, 'samples': 28361088, 'steps': 147713, 'loss/train': 1.488633155822754}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:17 - INFO - __main__ - Step 147714: {'lr': 2.945325941059729e-07, 'samples': 28361088, 'steps': 147713, 'loss/train': 1.488633155822754}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:17 - INFO - __main__ - Step 147714: {'lr': 2.945325941059729e-07, 'samples': 28361088, 'steps': 147713, 'loss/train': 1.488633155822754}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:23 - INFO - __main__ - Step 147725: {'lr': 2.91706675157849e-07, 'samples': 28363200, 'steps': 147724, 'loss/train': 1.3511896133422852}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:25 - INFO - __main__ - Step 147730: {'lr': 2.904266671239775e-07, 'samples': 28364160, 'steps': 147729, 'loss/train': 1.3557158708572388}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:28 - INFO - __main__ - Step 147735: {'lr': 2.8914947197208884e-07, 'samples': 28365120, 'steps': 147734, 'loss/train': 1.3744174242019653}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:28 - INFO - __main__ - Step 147735: {'lr': 2.8914947197208884e-07, 'samples': 28365120, 'steps': 147734, 'loss/train': 1.3744174242019653}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:31 - INFO - __main__ - Step 147742: {'lr': 2.873661244286452e-07, 'samples': 28366464, 'steps': 147741, 'loss/train': 1.7972146272659302}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:33 - INFO - __main__ - Step 147746: {'lr': 2.86349544053488e-07, 'samples': 28367232, 'steps': 147745, 'loss/train': 0.9200687408447266}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:35 - INFO - __main__ - Step 147750: {'lr': 2.85334763952233e-07, 'samples': 28368000, 'steps': 147749, 'loss/train': 1.105513095855713}}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:37 - INFO - __main__ - Step 147755: {'lr': 2.840688204719666e-07, 'samples': 28368960, 'steps': 147754, 'loss/train': 1.1104148626327515}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:39 - INFO - __main__ - Step 147759: {'lr': 2.8305809101364375e-07, 'samples': 28369728, 'steps': 147758, 'loss/train': 1.7326353788375854}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:42 - INFO - __main__ - Step 147763: {'lr': 2.820491618529819e-07, 'samples': 28370496, 'steps': 147762, 'loss/train': 1.5299278497695923}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:43 - INFO - __main__ - Step 147767: {'lr': 2.810420329971697e-07, 'samples': 28371264, 'steps': 147766, 'loss/train': 1.4579064846038818}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:45 - INFO - __main__ - Step 147771: {'lr': 2.800367044535068e-07, 'samples': 28372032, 'steps': 147770, 'loss/train': 1.2091176509857178}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:48 - INFO - __main__ - Step 147776: {'lr': 2.787825754739237e-07, 'samples': 28372992, 'steps': 147775, 'loss/train': 0.867586076259613}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:50 - INFO - __main__ - Step 147780: {'lr': 2.777812976590688e-07, 'samples': 28373760, 'steps': 147779, 'loss/train': 0.8709926605224609}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:52 - INFO - __main__ - Step 147784: {'lr': 2.767818201798444e-07, 'samples': 28374528, 'steps': 147783, 'loss/train': 0.9401568174362183}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:52 - INFO - __main__ - Step 147784: {'lr': 2.767818201798444e-07, 'samples': 28374528, 'steps': 147783, 'loss/train': 0.9401568174362183}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:55 - INFO - __main__ - Step 147791: {'lr': 2.750370666705637e-07, 'samples': 28375872, 'steps': 147790, 'loss/train': 4.566514015197754}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:01:58 - INFO - __main__ - Step 147797: {'lr': 2.735459520275263e-07, 'samples': 28377024, 'steps': 147796, 'loss/train': 1.2579351663589478}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:00 - INFO - __main__ - Step 147801: {'lr': 2.7255412605506347e-07, 'samples': 28377792, 'steps': 147800, 'loss/train': 1.8269883394241333}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:02 - INFO - __main__ - Step 147805: {'lr': 2.7156410045595103e-07, 'samples': 28378560, 'steps': 147804, 'loss/train': 1.4766504764556885}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:04 - INFO - __main__ - Step 147809: {'lr': 2.705758752372944e-07, 'samples': 28379328, 'steps': 147808, 'loss/train': 1.1265336275100708}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:05 - INFO - __main__ - Step 147813: {'lr': 2.695894504062546e-07, 'samples': 28380096, 'steps': 147812, 'loss/train': 1.171671748161316}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:07 - INFO - __main__ - Step 147817: {'lr': 2.6860482596993694e-07, 'samples': 28380864, 'steps': 147816, 'loss/train': 1.108141303062439}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:10 - INFO - __main__ - Step 147822: {'lr': 2.6737657724038025e-07, 'samples': 28381824, 'steps': 147821, 'loss/train': 1.0816516876220703}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:12 - INFO - __main__ - Step 147826: {'lr': 2.663960037180557e-07, 'samples': 28382592, 'steps': 147825, 'loss/train': 1.0448951721191406}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:12 - INFO - __main__ - Step 147826: {'lr': 2.663960037180557e-07, 'samples': 28382592, 'steps': 147825, 'loss/train': 1.0448951721191406}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:15 - INFO - __main__ - Step 147833: {'lr': 2.646843323134418e-07, 'samples': 28383936, 'steps': 147832, 'loss/train': 0.9213279485702515}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:17 - INFO - __main__ - Step 147838: {'lr': 2.6346508568575457e-07, 'samples': 28384896, 'steps': 147837, 'loss/train': 1.4709118604660034}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:17 - INFO - __main__ - Step 147838: {'lr': 2.6346508568575457e-07, 'samples': 28384896, 'steps': 147837, 'loss/train': 1.4709118604660034}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:22 - INFO - __main__ - Step 147846: {'lr': 2.615201425134606e-07, 'samples': 28386432, 'steps': 147845, 'loss/train': 0.8744367957115173}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:24 - INFO - __main__ - Step 147850: {'lr': 2.6055037160313543e-07, 'samples': 28387200, 'steps': 147849, 'loss/train': 0.8140342235565186}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:25 - INFO - __main__ - Step 147854: {'lr': 2.5958240115267486e-07, 'samples': 28387968, 'steps': 147853, 'loss/train': 1.0148130655288696}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:27 - INFO - __main__ - Step 147858: {'lr': 2.586162311690732e-07, 'samples': 28388736, 'steps': 147857, 'loss/train': 1.391814112663269}6}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:30 - INFO - __main__ - Step 147863: {'lr': 2.5741105060669844e-07, 'samples': 28389696, 'steps': 147862, 'loss/train': 1.3097550868988037}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:32 - INFO - __main__ - Step 147867: {'lr': 2.5644893169896045e-07, 'samples': 28390464, 'steps': 147866, 'loss/train': 1.7548794746398926}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:34 - INFO - __main__ - Step 147871: {'lr': 2.554886132807022e-07, 'samples': 28391232, 'steps': 147870, 'loss/train': 1.2795475721359253}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:35 - INFO - __main__ - Step 147875: {'lr': 2.5453009535877924e-07, 'samples': 28392000, 'steps': 147874, 'loss/train': 1.2156668901443481}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:38 - INFO - __main__ - Step 147880: {'lr': 2.5333447991490734e-07, 'samples': 28392960, 'steps': 147879, 'loss/train': 0.8160572052001953}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:40 - INFO - __main__ - Step 147884: {'lr': 2.523800131350729e-07, 'samples': 28393728, 'steps': 147883, 'loss/train': 1.358131766319275}3}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:40 - INFO - __main__ - Step 147884: {'lr': 2.523800131350729e-07, 'samples': 28393728, 'steps': 147883, 'loss/train': 1.358131766319275}3}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:43 - INFO - __main__ - Step 147891: {'lr': 2.507140287728027e-07, 'samples': 28395072, 'steps': 147890, 'loss/train': 1.0553126335144043}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:46 - INFO - __main__ - Step 147895: {'lr': 2.4976451343611264e-07, 'samples': 28395840, 'steps': 147894, 'loss/train': 0.9764859676361084}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:48 - INFO - __main__ - Step 147900: {'lr': 2.485801512721042e-07, 'samples': 28396800, 'steps': 147899, 'loss/train': 0.5887227654457092}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:48 - INFO - __main__ - Step 147900: {'lr': 2.485801512721042e-07, 'samples': 28396800, 'steps': 147899, 'loss/train': 0.5887227654457092}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:52 - INFO - __main__ - Step 147907: {'lr': 2.469267706791956e-07, 'samples': 28398144, 'steps': 147906, 'loss/train': 1.4615585803985596}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:54 - INFO - __main__ - Step 147911: {'lr': 2.459844575339876e-07, 'samples': 28398912, 'steps': 147910, 'loss/train': 1.476236343383789}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:56 - INFO - __main__ - Step 147915: {'lr': 2.450439449536712e-07, 'samples': 28399680, 'steps': 147914, 'loss/train': 1.3009071350097656}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:58 - INFO - __main__ - Step 147920: {'lr': 2.438708362829667e-07, 'samples': 28400640, 'steps': 147919, 'loss/train': 1.3192414045333862}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:58 - INFO - __main__ - Step 147920: {'lr': 2.438708362829667e-07, 'samples': 28400640, 'steps': 147919, 'loss/train': 1.3192414045333862}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:02:58 - INFO - __main__ - Step 147920: {'lr': 2.438708362829667e-07, 'samples': 28400640, 'steps': 147919, 'loss/train': 1.3192414045333862}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:03 - INFO - __main__ - Step 147931: {'lr': 2.4129990041668603e-07, 'samples': 28402752, 'steps': 147930, 'loss/train': 1.2874237298965454}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:06 - INFO - __main__ - Step 147936: {'lr': 2.4013579469295275e-07, 'samples': 28403712, 'steps': 147935, 'loss/train': 1.1103729009628296}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:08 - INFO - __main__ - Step 147940: {'lr': 2.3920653579628806e-07, 'samples': 28404480, 'steps': 147939, 'loss/train': 1.4278554916381836}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:10 - INFO - __main__ - Step 147944: {'lr': 2.3827907751336475e-07, 'samples': 28405248, 'steps': 147943, 'loss/train': 1.1298106908798218}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:12 - INFO - __main__ - Step 147948: {'lr': 2.3735341985089976e-07, 'samples': 28406016, 'steps': 147947, 'loss/train': 1.4128119945526123}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:13 - INFO - __main__ - Step 147952: {'lr': 2.3642956281552664e-07, 'samples': 28406784, 'steps': 147951, 'loss/train': 0.9749755263328552}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:15 - INFO - __main__ - Step 147956: {'lr': 2.3550750641393447e-07, 'samples': 28407552, 'steps': 147955, 'loss/train': 1.2197058200836182}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:15 - INFO - __main__ - Step 147956: {'lr': 2.3550750641393447e-07, 'samples': 28407552, 'steps': 147955, 'loss/train': 1.2197058200836182}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:15 - INFO - __main__ - Step 147956: {'lr': 2.3550750641393447e-07, 'samples': 28407552, 'steps': 147955, 'loss/train': 1.2197058200836182}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:21 - INFO - __main__ - Step 147968: {'lr': 2.327521410781519e-07, 'samples': 28409856, 'steps': 147967, 'loss/train': 1.7823749780654907}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:23 - INFO - __main__ - Step 147972: {'lr': 2.3183728727799168e-07, 'samples': 28410624, 'steps': 147971, 'loss/train': 1.713012933731079}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:26 - INFO - __main__ - Step 147977: {'lr': 2.3069625221633362e-07, 'samples': 28411584, 'steps': 147976, 'loss/train': 1.1038331985473633}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:28 - INFO - __main__ - Step 147981: {'lr': 2.297854499259122e-07, 'samples': 28412352, 'steps': 147980, 'loss/train': 1.2281211614608765}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:30 - INFO - __main__ - Step 147985: {'lr': 2.288764483171224e-07, 'samples': 28413120, 'steps': 147984, 'loss/train': 1.147247314453125}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:30 - INFO - __main__ - Step 147985: {'lr': 2.288764483171224e-07, 'samples': 28413120, 'steps': 147984, 'loss/train': 1.147247314453125}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:33 - INFO - __main__ - Step 147992: {'lr': 2.2729002841176404e-07, 'samples': 28414464, 'steps': 147991, 'loss/train': 1.633750557899475}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:36 - INFO - __main__ - Step 147997: {'lr': 2.2616024764632315e-07, 'samples': 28415424, 'steps': 147996, 'loss/train': 1.4520940780639648}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:38 - INFO - __main__ - Step 148002: {'lr': 2.2503328048681383e-07, 'samples': 28416384, 'steps': 148001, 'loss/train': 1.1756608486175537}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:40 - INFO - __main__ - Step 148006: {'lr': 2.241337325640247e-07, 'samples': 28417152, 'steps': 148005, 'loss/train': 1.0472534894943237}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:40 - INFO - __main__ - Step 148006: {'lr': 2.241337325640247e-07, 'samples': 28417152, 'steps': 148005, 'loss/train': 1.0472534894943237}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:43 - INFO - __main__ - Step 148013: {'lr': 2.225638566913657e-07, 'samples': 28418496, 'steps': 148012, 'loss/train': 1.1129719018936157}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:46 - INFO - __main__ - Step 148018: {'lr': 2.2144589315606367e-07, 'samples': 28419456, 'steps': 148017, 'loss/train': 1.392519235610962}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:46 - INFO - __main__ - Step 148018: {'lr': 2.2144589315606367e-07, 'samples': 28419456, 'steps': 148017, 'loss/train': 1.392519235610962}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:50 - INFO - __main__ - Step 148026: {'lr': 2.196630039155989e-07, 'samples': 28420992, 'steps': 148025, 'loss/train': 1.6297005414962769}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:52 - INFO - __main__ - Step 148030: {'lr': 2.1877426042413162e-07, 'samples': 28421760, 'steps': 148029, 'loss/train': 1.2749385833740234}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:54 - INFO - __main__ - Step 148034: {'lr': 2.1788731769373238e-07, 'samples': 28422528, 'steps': 148033, 'loss/train': 1.7632861137390137}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:56 - INFO - __main__ - Step 148038: {'lr': 2.1700217573075721e-07, 'samples': 28423296, 'steps': 148037, 'loss/train': 1.2303986549377441}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:58 - INFO - __main__ - Step 148043: {'lr': 2.158982806159948e-07, 'samples': 28424256, 'steps': 148042, 'loss/train': 0.9601913094520569}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:03:58 - INFO - __main__ - Step 148043: {'lr': 2.158982806159948e-07, 'samples': 28424256, 'steps': 148042, 'loss/train': 0.9601913094520569}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:02 - INFO - __main__ - Step 148051: {'lr': 2.1413790097837837e-07, 'samples': 28425792, 'steps': 148050, 'loss/train': 1.215686321258545}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:04 - INFO - __main__ - Step 148055: {'lr': 2.13260412348093e-07, 'samples': 28426560, 'steps': 148054, 'loss/train': 0.9936357140541077}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:06 - INFO - __main__ - Step 148059: {'lr': 2.123847245186217e-07, 'samples': 28427328, 'steps': 148058, 'loss/train': 1.3282915353775024}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:08 - INFO - __main__ - Step 148064: {'lr': 2.112926471175558e-07, 'samples': 28428288, 'steps': 148063, 'loss/train': 0.5920829772949219}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:11 - INFO - __main__ - Step 148069: {'lr': 2.1020338348989864e-07, 'samples': 28429248, 'steps': 148068, 'loss/train': 1.3938019275665283}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:11 - INFO - __main__ - Step 148069: {'lr': 2.1020338348989864e-07, 'samples': 28429248, 'steps': 148068, 'loss/train': 1.3938019275665283}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:14 - INFO - __main__ - Step 148076: {'lr': 2.086831415738788e-07, 'samples': 28430592, 'steps': 148075, 'loss/train': 1.1307411193847656}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:16 - INFO - __main__ - Step 148080: {'lr': 2.07816908051911e-07, 'samples': 28431360, 'steps': 148079, 'loss/train': 1.6497728824615479}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:18 - INFO - __main__ - Step 148085: {'lr': 2.06736648581507e-07, 'samples': 28432320, 'steps': 148084, 'loss/train': 1.9081549644470215}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:18 - INFO - __main__ - Step 148085: {'lr': 2.06736648581507e-07, 'samples': 28432320, 'steps': 148084, 'loss/train': 1.9081549644470215}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:23 - INFO - __main__ - Step 148093: {'lr': 2.050140861893923e-07, 'samples': 28433856, 'steps': 148092, 'loss/train': 1.3308465480804443}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:25 - INFO - __main__ - Step 148097: {'lr': 2.041555062806122e-07, 'samples': 28434624, 'steps': 148096, 'loss/train': 1.0532768964767456}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:26 - INFO - __main__ - Step 148101: {'lr': 2.03298727238288e-07, 'samples': 28435392, 'steps': 148100, 'loss/train': 1.3590714931488037}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:26 - INFO - __main__ - Step 148101: {'lr': 2.03298727238288e-07, 'samples': 28435392, 'steps': 148100, 'loss/train': 1.3590714931488037}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:26 - INFO - __main__ - Step 148101: {'lr': 2.03298727238288e-07, 'samples': 28435392, 'steps': 148100, 'loss/train': 1.3590714931488037}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:32 - INFO - __main__ - Step 148111: {'lr': 2.011646584636484e-07, 'samples': 28437312, 'steps': 148110, 'loss/train': 1.2210217714309692}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:34 - INFO - __main__ - Step 148115: {'lr': 2.0031418250243684e-07, 'samples': 28438080, 'steps': 148114, 'loss/train': 0.6635581254959106}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:34 - INFO - __main__ - Step 148115: {'lr': 2.0031418250243684e-07, 'samples': 28438080, 'steps': 148114, 'loss/train': 0.6635581254959106}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:38 - INFO - __main__ - Step 148122: {'lr': 1.9883018297550681e-07, 'samples': 28439424, 'steps': 148121, 'loss/train': 0.8102813959121704}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:40 - INFO - __main__ - Step 148127: {'lr': 1.977735600080799e-07, 'samples': 28440384, 'steps': 148126, 'loss/train': 1.0826349258422852}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:42 - INFO - __main__ - Step 148131: {'lr': 1.9693028766010203e-07, 'samples': 28441152, 'steps': 148130, 'loss/train': 1.283203363418579}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:44 - INFO - __main__ - Step 148135: {'lr': 1.960888162306218e-07, 'samples': 28441920, 'steps': 148134, 'loss/train': 0.5120120644569397}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:46 - INFO - __main__ - Step 148139: {'lr': 1.9524914572577325e-07, 'samples': 28442688, 'steps': 148138, 'loss/train': 0.8983304500579834}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:46 - INFO - __main__ - Step 148139: {'lr': 1.9524914572577325e-07, 'samples': 28442688, 'steps': 148138, 'loss/train': 0.8983304500579834}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:50 - INFO - __main__ - Step 148145: {'lr': 1.9399301671535475e-07, 'samples': 28443840, 'steps': 148144, 'loss/train': 0.17297394573688507}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:52 - INFO - __main__ - Step 148150: {'lr': 1.9294933790425796e-07, 'samples': 28444800, 'steps': 148149, 'loss/train': 1.3218934535980225}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:55 - INFO - __main__ - Step 148154: {'lr': 1.9211642092045732e-07, 'samples': 28445568, 'steps': 148153, 'loss/train': 1.299556016921997}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:57 - INFO - __main__ - Step 148158: {'lr': 1.9128530488990435e-07, 'samples': 28446336, 'steps': 148157, 'loss/train': 0.8652459979057312}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:04:59 - INFO - __main__ - Step 148162: {'lr': 1.9045598981856648e-07, 'samples': 28447104, 'steps': 148161, 'loss/train': 0.8609744310379028}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:00 - INFO - __main__ - Step 148166: {'lr': 1.8962847571246666e-07, 'samples': 28447872, 'steps': 148165, 'loss/train': 0.993301272392273}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:03 - INFO - __main__ - Step 148171: {'lr': 1.8859661569622933e-07, 'samples': 28448832, 'steps': 148170, 'loss/train': 1.4992244243621826}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:05 - INFO - __main__ - Step 148175: {'lr': 1.8777315378362692e-07, 'samples': 28449600, 'steps': 148174, 'loss/train': 0.8013189435005188}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:05 - INFO - __main__ - Step 148175: {'lr': 1.8777315378362692e-07, 'samples': 28449600, 'steps': 148174, 'loss/train': 0.8013189435005188}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:08 - INFO - __main__ - Step 148182: {'lr': 1.8633642905924175e-07, 'samples': 28450944, 'steps': 148181, 'loss/train': 1.6723214387893677}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:10 - INFO - __main__ - Step 148187: {'lr': 1.8531357397685856e-07, 'samples': 28451904, 'steps': 148186, 'loss/train': 1.3668208122253418}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:12 - INFO - __main__ - Step 148191: {'lr': 1.844973160379515e-07, 'samples': 28452672, 'steps': 148190, 'loss/train': 1.3481011390686035}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:12 - INFO - __main__ - Step 148191: {'lr': 1.844973160379515e-07, 'samples': 28452672, 'steps': 148190, 'loss/train': 1.3481011390686035}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:16 - INFO - __main__ - Step 148199: {'lr': 1.828702031905216e-07, 'samples': 28454208, 'steps': 148198, 'loss/train': 1.2195147275924683}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:18 - INFO - __main__ - Step 148203: {'lr': 1.8205934829373937e-07, 'samples': 28454976, 'steps': 148202, 'loss/train': 1.5672129392623901}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:21 - INFO - __main__ - Step 148208: {'lr': 1.8104831236590014e-07, 'samples': 28455936, 'steps': 148207, 'loss/train': 1.0009839534759521}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:23 - INFO - __main__ - Step 148212: {'lr': 1.8024150978523946e-07, 'samples': 28456704, 'steps': 148211, 'loss/train': 1.3340163230895996}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:25 - INFO - __main__ - Step 148216: {'lr': 1.794365082434246e-07, 'samples': 28457472, 'steps': 148215, 'loss/train': 1.488923192024231}6}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:27 - INFO - __main__ - Step 148220: {'lr': 1.7863330774625652e-07, 'samples': 28458240, 'steps': 148219, 'loss/train': 1.2255045175552368}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:28 - INFO - __main__ - Step 148224: {'lr': 1.7783190829956385e-07, 'samples': 28459008, 'steps': 148223, 'loss/train': 1.1037113666534424}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:30 - INFO - __main__ - Step 148228: {'lr': 1.7703230990906427e-07, 'samples': 28459776, 'steps': 148227, 'loss/train': 2.2034521102905273}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:33 - INFO - __main__ - Step 148232: {'lr': 1.7623451258055868e-07, 'samples': 28460544, 'steps': 148231, 'loss/train': 1.323881983757019}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:33 - INFO - __main__ - Step 148232: {'lr': 1.7623451258055868e-07, 'samples': 28460544, 'steps': 148231, 'loss/train': 1.323881983757019}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:36 - INFO - __main__ - Step 148239: {'lr': 1.7484270107837043e-07, 'samples': 28461888, 'steps': 148238, 'loss/train': 1.461181879043579}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:38 - INFO - __main__ - Step 148244: {'lr': 1.738519270244321e-07, 'samples': 28462848, 'steps': 148243, 'loss/train': 1.549454689025879}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:41 - INFO - __main__ - Step 148249: {'lr': 1.7286396716564845e-07, 'samples': 28463808, 'steps': 148248, 'loss/train': 1.6633731126785278}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:41 - INFO - __main__ - Step 148249: {'lr': 1.7286396716564845e-07, 'samples': 28463808, 'steps': 148248, 'loss/train': 1.6633731126785278}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:44 - INFO - __main__ - Step 148256: {'lr': 1.714855512324476e-07, 'samples': 28465152, 'steps': 148255, 'loss/train': 0.9011018872261047}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:46 - INFO - __main__ - Step 148260: {'lr': 1.707003614981606e-07, 'samples': 28465920, 'steps': 148259, 'loss/train': 0.5996752381324768}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:48 - INFO - __main__ - Step 148264: {'lr': 1.6991697287152552e-07, 'samples': 28466688, 'steps': 148263, 'loss/train': 1.5107617378234863}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:48 - INFO - __main__ - Step 148264: {'lr': 1.6991697287152552e-07, 'samples': 28466688, 'steps': 148263, 'loss/train': 1.5107617378234863}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:48 - INFO - __main__ - Step 148264: {'lr': 1.6991697287152552e-07, 'samples': 28466688, 'steps': 148263, 'loss/train': 1.5107617378234863}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:54 - INFO - __main__ - Step 148276: {'lr': 1.6757761369384783e-07, 'samples': 28468992, 'steps': 148275, 'loss/train': 1.5367859601974487}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:56 - INFO - __main__ - Step 148281: {'lr': 1.666076649463888e-07, 'samples': 28469952, 'steps': 148280, 'loss/train': 1.2619932889938354}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:56 - INFO - __main__ - Step 148281: {'lr': 1.666076649463888e-07, 'samples': 28469952, 'steps': 148280, 'loss/train': 1.2619932889938354}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:05:56 - INFO - __main__ - Step 148281: {'lr': 1.666076649463888e-07, 'samples': 28469952, 'steps': 148280, 'loss/train': 1.2619932889938354}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:02 - INFO - __main__ - Step 148292: {'lr': 1.6448368397162238e-07, 'samples': 28472064, 'steps': 148291, 'loss/train': 1.1561568975448608}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:05 - INFO - __main__ - Step 148298: {'lr': 1.6333089006054236e-07, 'samples': 28473216, 'steps': 148297, 'loss/train': 1.2342702150344849}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:07 - INFO - __main__ - Step 148302: {'lr': 1.6256461223629336e-07, 'samples': 28473984, 'steps': 148301, 'loss/train': 0.7772263884544373}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:09 - INFO - __main__ - Step 148306: {'lr': 1.6180013557831608e-07, 'samples': 28474752, 'steps': 148305, 'loss/train': 1.0246626138687134}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:11 - INFO - __main__ - Step 148310: {'lr': 1.6103746009216157e-07, 'samples': 28475520, 'steps': 148309, 'loss/train': 1.3202219009399414}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:12 - INFO - __main__ - Step 148314: {'lr': 1.6027658578329774e-07, 'samples': 28476288, 'steps': 148313, 'loss/train': 1.434384822845459}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:15 - INFO - __main__ - Step 148318: {'lr': 1.5951751265722013e-07, 'samples': 28477056, 'steps': 148317, 'loss/train': 1.6695586442947388}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:15 - INFO - __main__ - Step 148318: {'lr': 1.5951751265722013e-07, 'samples': 28477056, 'steps': 148317, 'loss/train': 1.6695586442947388}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:19 - INFO - __main__ - Step 148326: {'lr': 1.5800476997529512e-07, 'samples': 28478592, 'steps': 148325, 'loss/train': 1.07534658908844}8}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:20 - INFO - __main__ - Step 148330: {'lr': 1.5725110043035563e-07, 'samples': 28479360, 'steps': 148329, 'loss/train': 1.1531052589416504}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:22 - INFO - __main__ - Step 148334: {'lr': 1.564992320899905e-07, 'samples': 28480128, 'steps': 148333, 'loss/train': 1.2736127376556396}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:25 - INFO - __main__ - Step 148339: {'lr': 1.555619296167987e-07, 'samples': 28481088, 'steps': 148338, 'loss/train': 1.2668399810791016}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:25 - INFO - __main__ - Step 148339: {'lr': 1.555619296167987e-07, 'samples': 28481088, 'steps': 148338, 'loss/train': 1.2668399810791016}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:29 - INFO - __main__ - Step 148347: {'lr': 1.5406809961854283e-07, 'samples': 28482624, 'steps': 148346, 'loss/train': 1.3248862028121948}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:30 - INFO - __main__ - Step 148351: {'lr': 1.533238864580566e-07, 'samples': 28483392, 'steps': 148350, 'loss/train': 1.0418822765350342}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:32 - INFO - __main__ - Step 148355: {'lr': 1.5258147453045545e-07, 'samples': 28484160, 'steps': 148354, 'loss/train': 1.4250106811523438}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:32 - INFO - __main__ - Step 148355: {'lr': 1.5258147453045545e-07, 'samples': 28484160, 'steps': 148354, 'loss/train': 1.4250106811523438}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:37 - INFO - __main__ - Step 148363: {'lr': 1.5110205439541892e-07, 'samples': 28485696, 'steps': 148362, 'loss/train': 1.2237542867660522}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:39 - INFO - __main__ - Step 148367: {'lr': 1.5036504619861392e-07, 'samples': 28486464, 'steps': 148366, 'loss/train': 1.2951220273971558}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:40 - INFO - __main__ - Step 148371: {'lr': 1.4962983925606576e-07, 'samples': 28487232, 'steps': 148370, 'loss/train': 0.8507276773452759}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:42 - INFO - __main__ - Step 148375: {'lr': 1.4889643357304804e-07, 'samples': 28488000, 'steps': 148374, 'loss/train': 0.6225084066390991}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:45 - INFO - __main__ - Step 148380: {'lr': 1.4798220949854834e-07, 'samples': 28488960, 'steps': 148379, 'loss/train': 1.3567720651626587}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:47 - INFO - __main__ - Step 148384: {'lr': 1.4725285666883358e-07, 'samples': 28489728, 'steps': 148383, 'loss/train': 0.9547638297080994}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:49 - INFO - __main__ - Step 148388: {'lr': 1.4652530511577444e-07, 'samples': 28490496, 'steps': 148387, 'loss/train': 1.196874737739563}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:50 - INFO - __main__ - Step 148392: {'lr': 1.457995548446167e-07, 'samples': 28491264, 'steps': 148391, 'loss/train': 1.403191089630127}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:52 - INFO - __main__ - Step 148396: {'lr': 1.450756058606617e-07, 'samples': 28492032, 'steps': 148395, 'loss/train': 1.1351863145828247}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:55 - INFO - __main__ - Step 148401: {'lr': 1.4417320269868862e-07, 'samples': 28492992, 'steps': 148400, 'loss/train': 1.1175259351730347}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:55 - INFO - __main__ - Step 148401: {'lr': 1.4417320269868862e-07, 'samples': 28492992, 'steps': 148400, 'loss/train': 1.1175259351730347}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:06:59 - INFO - __main__ - Step 148408: {'lr': 1.4291456668374947e-07, 'samples': 28494336, 'steps': 148407, 'loss/train': 1.601609706878662}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:00 - INFO - __main__ - Step 148412: {'lr': 1.421978229004528e-07, 'samples': 28495104, 'steps': 148411, 'loss/train': 1.281636118888855}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:02 - INFO - __main__ - Step 148416: {'lr': 1.4148288043028256e-07, 'samples': 28495872, 'steps': 148415, 'loss/train': 1.19974946975708}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:05 - INFO - __main__ - Step 148421: {'lr': 1.4059173544697657e-07, 'samples': 28496832, 'steps': 148420, 'loss/train': 0.8852680921554565}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:07 - INFO - __main__ - Step 148425: {'lr': 1.3988084595015217e-07, 'samples': 28497600, 'steps': 148424, 'loss/train': 1.3240045309066772}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:09 - INFO - __main__ - Step 148429: {'lr': 1.3917175778313529e-07, 'samples': 28498368, 'steps': 148428, 'loss/train': 1.1014258861541748}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:10 - INFO - __main__ - Step 148433: {'lr': 1.3846447095106074e-07, 'samples': 28499136, 'steps': 148432, 'loss/train': 0.8909156918525696}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:12 - INFO - __main__ - Step 148437: {'lr': 1.3775898545903554e-07, 'samples': 28499904, 'steps': 148436, 'loss/train': 1.0902682542800903}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:15 - INFO - __main__ - Step 148442: {'lr': 1.3687966173617228e-07, 'samples': 28500864, 'steps': 148441, 'loss/train': 1.280573844909668}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:15 - INFO - __main__ - Step 148442: {'lr': 1.3687966173617228e-07, 'samples': 28500864, 'steps': 148441, 'loss/train': 1.280573844909668}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:18 - INFO - __main__ - Step 148449: {'lr': 1.3565333707393767e-07, 'samples': 28502208, 'steps': 148448, 'loss/train': 1.4798576831817627}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:20 - INFO - __main__ - Step 148453: {'lr': 1.3495505699279154e-07, 'samples': 28502976, 'steps': 148452, 'loss/train': 1.6019643545150757}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:20 - INFO - __main__ - Step 148453: {'lr': 1.3495505699279154e-07, 'samples': 28502976, 'steps': 148452, 'loss/train': 1.6019643545150757}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:24 - INFO - __main__ - Step 148461: {'lr': 1.3356390093149928e-07, 'samples': 28504512, 'steps': 148460, 'loss/train': 1.7491270303726196}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:26 - INFO - __main__ - Step 148465: {'lr': 1.328710249614007e-07, 'samples': 28505280, 'steps': 148464, 'loss/train': 1.492993950843811}6}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:29 - INFO - __main__ - Step 148470: {'lr': 1.3200746319050328e-07, 'samples': 28506240, 'steps': 148469, 'loss/train': 1.0321356058120728}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:31 - INFO - __main__ - Step 148475: {'lr': 1.311467160861335e-07, 'samples': 28507200, 'steps': 148474, 'loss/train': 0.7179288864135742}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:31 - INFO - __main__ - Step 148475: {'lr': 1.311467160861335e-07, 'samples': 28507200, 'steps': 148474, 'loss/train': 0.7179288864135742}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:35 - INFO - __main__ - Step 148482: {'lr': 1.2994639879826408e-07, 'samples': 28508544, 'steps': 148481, 'loss/train': 1.113702654838562}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:36 - INFO - __main__ - Step 148486: {'lr': 1.2926298013049563e-07, 'samples': 28509312, 'steps': 148485, 'loss/train': 1.1540107727050781}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:39 - INFO - __main__ - Step 148491: {'lr': 1.2841124002405112e-07, 'samples': 28510272, 'steps': 148490, 'loss/train': 1.3165984153747559}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:41 - INFO - __main__ - Step 148495: {'lr': 1.2773187452752578e-07, 'samples': 28511040, 'steps': 148494, 'loss/train': 1.1529512405395508}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:41 - INFO - __main__ - Step 148495: {'lr': 1.2773187452752578e-07, 'samples': 28511040, 'steps': 148494, 'loss/train': 1.1529512405395508}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:44 - INFO - __main__ - Step 148501: {'lr': 1.267162039418901e-07, 'samples': 28512192, 'steps': 148500, 'loss/train': 1.598413109779358}8}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:47 - INFO - __main__ - Step 148507: {'lr': 1.257045865619144e-07, 'samples': 28513344, 'steps': 148506, 'loss/train': 1.5483105182647705}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:47 - INFO - __main__ - Step 148507: {'lr': 1.257045865619144e-07, 'samples': 28513344, 'steps': 148506, 'loss/train': 1.5483105182647705}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:51 - INFO - __main__ - Step 148514: {'lr': 1.245294891087656e-07, 'samples': 28514688, 'steps': 148513, 'loss/train': 0.3149808943271637}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:52 - INFO - __main__ - Step 148518: {'lr': 1.2386048182883026e-07, 'samples': 28515456, 'steps': 148517, 'loss/train': 1.0768243074417114}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:55 - INFO - __main__ - Step 148522: {'lr': 1.2319327599422115e-07, 'samples': 28516224, 'steps': 148521, 'loss/train': 1.4521968364715576}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:57 - INFO - __main__ - Step 148527: {'lr': 1.2236180199076686e-07, 'samples': 28517184, 'steps': 148526, 'loss/train': 0.8851209878921509}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:59 - INFO - __main__ - Step 148532: {'lr': 1.2153314276250326e-07, 'samples': 28518144, 'steps': 148531, 'loss/train': 1.117923617362976}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:07:59 - INFO - __main__ - Step 148532: {'lr': 1.2153314276250326e-07, 'samples': 28518144, 'steps': 148531, 'loss/train': 1.117923617362976}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:03 - INFO - __main__ - Step 148539: {'lr': 1.2037774868306485e-07, 'samples': 28519488, 'steps': 148538, 'loss/train': 1.5785672664642334}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:05 - INFO - __main__ - Step 148543: {'lr': 1.1972000051499454e-07, 'samples': 28520256, 'steps': 148542, 'loss/train': 1.4898360967636108}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:07 - INFO - __main__ - Step 148548: {'lr': 1.189003486299356e-07, 'samples': 28521216, 'steps': 148547, 'loss/train': 1.5293558835983276}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:07 - INFO - __main__ - Step 148548: {'lr': 1.189003486299356e-07, 'samples': 28521216, 'steps': 148547, 'loss/train': 1.5293558835983276}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:11 - INFO - __main__ - Step 148556: {'lr': 1.1759476043118889e-07, 'samples': 28522752, 'steps': 148555, 'loss/train': 1.1717798709869385}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:13 - INFO - __main__ - Step 148560: {'lr': 1.169446685652109e-07, 'samples': 28523520, 'steps': 148559, 'loss/train': 0.9456471800804138}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:15 - INFO - __main__ - Step 148564: {'lr': 1.1629637819438044e-07, 'samples': 28524288, 'steps': 148563, 'loss/train': 0.9589516520500183}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:17 - INFO - __main__ - Step 148569: {'lr': 1.1548854859058655e-07, 'samples': 28525248, 'steps': 148568, 'loss/train': 0.9858706593513489}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:17 - INFO - __main__ - Step 148569: {'lr': 1.1548854859058655e-07, 'samples': 28525248, 'steps': 148568, 'loss/train': 0.9858706593513489}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:22 - INFO - __main__ - Step 148577: {'lr': 1.1420187612173893e-07, 'samples': 28526784, 'steps': 148576, 'loss/train': 1.6526533365249634}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:24 - INFO - __main__ - Step 148581: {'lr': 1.1356124215744501e-07, 'samples': 28527552, 'steps': 148580, 'loss/train': 1.234092116355896}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:25 - INFO - __main__ - Step 148585: {'lr': 1.1292240971269574e-07, 'samples': 28528320, 'steps': 148584, 'loss/train': 1.3224709033966064}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:25 - INFO - __main__ - Step 148585: {'lr': 1.1292240971269574e-07, 'samples': 28528320, 'steps': 148584, 'loss/train': 1.3224709033966064}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:28 - INFO - __main__ - Step 148590: {'lr': 1.1212640255070516e-07, 'samples': 28529280, 'steps': 148589, 'loss/train': 1.2780572175979614}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:30 - INFO - __main__ - Step 148596: {'lr': 1.1117490961251586e-07, 'samples': 28530432, 'steps': 148595, 'loss/train': 1.3866816759109497}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:32 - INFO - __main__ - Step 148600: {'lr': 1.1054283290709566e-07, 'samples': 28531200, 'steps': 148599, 'loss/train': 1.5202064514160156}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:35 - INFO - __main__ - Step 148605: {'lr': 1.0975527044335132e-07, 'samples': 28532160, 'steps': 148604, 'loss/train': 1.1778842210769653}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:37 - INFO - __main__ - Step 148610: {'lr': 1.0897052289679521e-07, 'samples': 28533120, 'steps': 148609, 'loss/train': 1.4142839908599854}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:39 - INFO - __main__ - Step 148614: {'lr': 1.0834475160589885e-07, 'samples': 28533888, 'steps': 148613, 'loss/train': 0.6181808114051819}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:39 - INFO - __main__ - Step 148614: {'lr': 1.0834475160589885e-07, 'samples': 28533888, 'steps': 148613, 'loss/train': 0.6181808114051819}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:43 - INFO - __main__ - Step 148621: {'lr': 1.0725398684646725e-07, 'samples': 28535232, 'steps': 148620, 'loss/train': 1.285168170928955}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:45 - INFO - __main__ - Step 148626: {'lr': 1.064782470941994e-07, 'samples': 28536192, 'steps': 148625, 'loss/train': 1.6266535520553589}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:47 - INFO - __main__ - Step 148631: {'lr': 1.0570532229606244e-07, 'samples': 28537152, 'steps': 148630, 'loss/train': 0.7557640671730042}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:47 - INFO - __main__ - Step 148631: {'lr': 1.0570532229606244e-07, 'samples': 28537152, 'steps': 148630, 'loss/train': 0.7557640671730042}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:51 - INFO - __main__ - Step 148638: {'lr': 1.0462795671817294e-07, 'samples': 28538496, 'steps': 148637, 'loss/train': 1.5654425621032715}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:53 - INFO - __main__ - Step 148642: {'lr': 1.0401479642152545e-07, 'samples': 28539264, 'steps': 148641, 'loss/train': 1.7181527614593506}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:55 - INFO - __main__ - Step 148647: {'lr': 1.0325087953494406e-07, 'samples': 28540224, 'steps': 148646, 'loss/train': 1.1078057289123535}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:55 - INFO - __main__ - Step 148647: {'lr': 1.0325087953494406e-07, 'samples': 28540224, 'steps': 148646, 'loss/train': 1.1078057289123535}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:08:55 - INFO - __main__ - Step 148647: {'lr': 1.0325087953494406e-07, 'samples': 28540224, 'steps': 148646, 'loss/train': 1.1078057289123535}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:01 - INFO - __main__ - Step 148658: {'lr': 1.0158017116293494e-07, 'samples': 28542336, 'steps': 148657, 'loss/train': 1.1537283658981323}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:03 - INFO - __main__ - Step 148663: {'lr': 1.0082526227597755e-07, 'samples': 28543296, 'steps': 148662, 'loss/train': 1.6956591606140137}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:06 - INFO - __main__ - Step 148668: {'lr': 1.0007316840682234e-07, 'samples': 28544256, 'steps': 148667, 'loss/train': 1.4858282804489136}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:06 - INFO - __main__ - Step 148668: {'lr': 1.0007316840682234e-07, 'samples': 28544256, 'steps': 148667, 'loss/train': 1.4858282804489136}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:09 - INFO - __main__ - Step 148675: {'lr': 9.902496623601098e-08, 'samples': 28545600, 'steps': 148674, 'loss/train': 1.2965996265411377}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:11 - INFO - __main__ - Step 148679: {'lr': 9.84284707990224e-08, 'samples': 28546368, 'steps': 148678, 'loss/train': 0.8088366389274597}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:14 - INFO - __main__ - Step 148684: {'lr': 9.768538504356661e-08, 'samples': 28547328, 'steps': 148683, 'loss/train': 1.9089263677597046}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:16 - INFO - __main__ - Step 148688: {'lr': 9.709294327708839e-08, 'samples': 28548096, 'steps': 148687, 'loss/train': 1.154142141342163}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:18 - INFO - __main__ - Step 148692: {'lr': 9.650230314892094e-08, 'samples': 28548864, 'steps': 148691, 'loss/train': 0.9130756855010986}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:19 - INFO - __main__ - Step 148696: {'lr': 9.591346466331086e-08, 'samples': 28549632, 'steps': 148695, 'loss/train': 1.305224895477295}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:21 - INFO - __main__ - Step 148700: {'lr': 9.532642782450473e-08, 'samples': 28550400, 'steps': 148699, 'loss/train': 0.9226098656654358}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:24 - INFO - __main__ - Step 148705: {'lr': 9.459516534823798e-08, 'samples': 28551360, 'steps': 148704, 'loss/train': 1.3714977502822876}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:26 - INFO - __main__ - Step 148709: {'lr': 9.401218223017982e-08, 'samples': 28552128, 'steps': 148708, 'loss/train': 1.105061650276184}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:28 - INFO - __main__ - Step 148713: {'lr': 9.343100077263689e-08, 'samples': 28552896, 'steps': 148712, 'loss/train': 1.130031704902649}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:29 - INFO - __main__ - Step 148717: {'lr': 9.285162097977251e-08, 'samples': 28553664, 'steps': 148716, 'loss/train': 0.7596377730369568}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:31 - INFO - __main__ - Step 148721: {'lr': 9.227404285580554e-08, 'samples': 28554432, 'steps': 148720, 'loss/train': 0.9187221527099609}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:34 - INFO - __main__ - Step 148726: {'lr': 9.155460380402824e-08, 'samples': 28555392, 'steps': 148725, 'loss/train': 1.1425954103469849}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:36 - INFO - __main__ - Step 148730: {'lr': 9.098107945021417e-08, 'samples': 28556160, 'steps': 148729, 'loss/train': 1.2909557819366455}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:36 - INFO - __main__ - Step 148730: {'lr': 9.098107945021417e-08, 'samples': 28556160, 'steps': 148729, 'loss/train': 1.2909557819366455}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:39 - INFO - __main__ - Step 148737: {'lr': 8.998174713173435e-08, 'samples': 28557504, 'steps': 148736, 'loss/train': 0.8240156769752502}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:41 - INFO - __main__ - Step 148741: {'lr': 8.941317741431587e-08, 'samples': 28558272, 'steps': 148740, 'loss/train': 1.908067226409912}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:43 - INFO - __main__ - Step 148745: {'lr': 8.884640939058053e-08, 'samples': 28559040, 'steps': 148744, 'loss/train': 1.3026037216186523}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:46 - INFO - __main__ - Step 148749: {'lr': 8.828144306460839e-08, 'samples': 28559808, 'steps': 148748, 'loss/train': 1.1787524223327637}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:47 - INFO - __main__ - Step 148753: {'lr': 8.771827844042401e-08, 'samples': 28560576, 'steps': 148752, 'loss/train': 1.415947675704956}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:49 - INFO - __main__ - Step 148757: {'lr': 8.715691552216298e-08, 'samples': 28561344, 'steps': 148756, 'loss/train': 1.3713430166244507}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:49 - INFO - __main__ - Step 148757: {'lr': 8.715691552216298e-08, 'samples': 28561344, 'steps': 148756, 'loss/train': 1.3713430166244507}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:53 - INFO - __main__ - Step 148765: {'lr': 8.603959481942591e-08, 'samples': 28562880, 'steps': 148764, 'loss/train': 0.34206968545913696}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:55 - INFO - __main__ - Step 148769: {'lr': 8.548363704302675e-08, 'samples': 28563648, 'steps': 148768, 'loss/train': 0.6129887700080872}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:09:57 - INFO - __main__ - Step 148774: {'lr': 8.479122349452917e-08, 'samples': 28564608, 'steps': 148773, 'loss/train': 1.0546016693115234}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:00 - INFO - __main__ - Step 148778: {'lr': 8.423931959822273e-08, 'samples': 28565376, 'steps': 148777, 'loss/train': 1.1762447357177734}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:02 - INFO - __main__ - Step 148782: {'lr': 8.368921743290292e-08, 'samples': 28566144, 'steps': 148781, 'loss/train': 1.497905969619751}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:02 - INFO - __main__ - Step 148782: {'lr': 8.368921743290292e-08, 'samples': 28566144, 'steps': 148781, 'loss/train': 1.497905969619751}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:05 - INFO - __main__ - Step 148789: {'lr': 8.273087407062452e-08, 'samples': 28567488, 'steps': 148788, 'loss/train': 0.8442835211753845}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:08 - INFO - __main__ - Step 148795: {'lr': 8.191382864788932e-08, 'samples': 28568640, 'steps': 148794, 'loss/train': 0.9032496809959412}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:08 - INFO - __main__ - Step 148795: {'lr': 8.191382864788932e-08, 'samples': 28568640, 'steps': 148794, 'loss/train': 0.9032496809959412}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:12 - INFO - __main__ - Step 148802: {'lr': 8.09657327090485e-08, 'samples': 28569984, 'steps': 148801, 'loss/train': 1.2436658143997192}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:13 - INFO - __main__ - Step 148806: {'lr': 8.042644101233232e-08, 'samples': 28570752, 'steps': 148805, 'loss/train': 0.6720967292785645}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:15 - INFO - __main__ - Step 148810: {'lr': 7.988895107405302e-08, 'samples': 28571520, 'steps': 148809, 'loss/train': 1.461071491241455}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:18 - INFO - __main__ - Step 148815: {'lr': 7.921962237994462e-08, 'samples': 28572480, 'steps': 148814, 'loss/train': 1.2191417217254639}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:18 - INFO - __main__ - Step 148815: {'lr': 7.921962237994462e-08, 'samples': 28572480, 'steps': 148814, 'loss/train': 1.2191417217254639}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:21 - INFO - __main__ - Step 148822: {'lr': 7.82872918486388e-08, 'samples': 28573824, 'steps': 148821, 'loss/train': 1.201392412185669}9}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:23 - INFO - __main__ - Step 148827: {'lr': 7.762471979397279e-08, 'samples': 28574784, 'steps': 148826, 'loss/train': 0.8880283832550049}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:23 - INFO - __main__ - Step 148827: {'lr': 7.762471979397279e-08, 'samples': 28574784, 'steps': 148826, 'loss/train': 0.8880283832550049}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:28 - INFO - __main__ - Step 148835: {'lr': 7.657046029019288e-08, 'samples': 28576320, 'steps': 148834, 'loss/train': 1.455342411994934}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:30 - INFO - __main__ - Step 148839: {'lr': 7.604603321575976e-08, 'samples': 28577088, 'steps': 148838, 'loss/train': 1.4976134300231934}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:31 - INFO - __main__ - Step 148843: {'lr': 7.552340793140488e-08, 'samples': 28577856, 'steps': 148842, 'loss/train': 1.4962459802627563}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:34 - INFO - __main__ - Step 148848: {'lr': 7.487266009889582e-08, 'samples': 28578816, 'steps': 148847, 'loss/train': 1.6581391096115112}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:36 - INFO - __main__ - Step 148852: {'lr': 7.435408885583806e-08, 'samples': 28579584, 'steps': 148851, 'loss/train': 1.1153931617736816}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:38 - INFO - __main__ - Step 148856: {'lr': 7.383731941501549e-08, 'samples': 28580352, 'steps': 148855, 'loss/train': 0.5957202911376953}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:40 - INFO - __main__ - Step 148860: {'lr': 7.332235178014735e-08, 'samples': 28581120, 'steps': 148859, 'loss/train': 1.0144908428192139}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:41 - INFO - __main__ - Step 148864: {'lr': 7.28091859549529e-08, 'samples': 28581888, 'steps': 148863, 'loss/train': 1.2689237594604492}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:44 - INFO - __main__ - Step 148869: {'lr': 7.217026247396952e-08, 'samples': 28582848, 'steps': 148868, 'loss/train': 1.2660002708435059}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:46 - INFO - __main__ - Step 148873: {'lr': 7.166115073400925e-08, 'samples': 28583616, 'steps': 148872, 'loss/train': 2.008755922317505}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:48 - INFO - __main__ - Step 148877: {'lr': 7.115384081574083e-08, 'samples': 28584384, 'steps': 148876, 'loss/train': 1.1886417865753174}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:48 - INFO - __main__ - Step 148877: {'lr': 7.115384081574083e-08, 'samples': 28584384, 'steps': 148876, 'loss/train': 1.1886417865753174}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:51 - INFO - __main__ - Step 148884: {'lr': 7.027038410306608e-08, 'samples': 28585728, 'steps': 148883, 'loss/train': 1.0403189659118652}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:51 - INFO - __main__ - Step 148884: {'lr': 7.027038410306608e-08, 'samples': 28585728, 'steps': 148883, 'loss/train': 1.0403189659118652}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:56 - INFO - __main__ - Step 148893: {'lr': 6.91426194319722e-08, 'samples': 28587456, 'steps': 148892, 'loss/train': 1.6087892055511475}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:58 - INFO - __main__ - Step 148897: {'lr': 6.864431867650844e-08, 'samples': 28588224, 'steps': 148896, 'loss/train': 1.3648629188537598}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:10:59 - INFO - __main__ - Step 148901: {'lr': 6.81478197644414e-08, 'samples': 28588992, 'steps': 148900, 'loss/train': 1.38480544090271}98}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:02 - INFO - __main__ - Step 148906: {'lr': 6.752972997209338e-08, 'samples': 28589952, 'steps': 148905, 'loss/train': 1.382609486579895}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:04 - INFO - __main__ - Step 148911: {'lr': 6.691445557258091e-08, 'samples': 28590912, 'steps': 148910, 'loss/train': 1.4177390336990356}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:04 - INFO - __main__ - Step 148911: {'lr': 6.691445557258091e-08, 'samples': 28590912, 'steps': 148910, 'loss/train': 1.4177390336990356}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:04 - INFO - __main__ - Step 148911: {'lr': 6.691445557258091e-08, 'samples': 28590912, 'steps': 148910, 'loss/train': 1.4177390336990356}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:09 - INFO - __main__ - Step 148921: {'lr': 6.569235297984588e-08, 'samples': 28592832, 'steps': 148920, 'loss/train': 1.4554566144943237}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:12 - INFO - __main__ - Step 148926: {'lr': 6.508552480036234e-08, 'samples': 28593792, 'steps': 148925, 'loss/train': 1.733709454536438}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:14 - INFO - __main__ - Step 148930: {'lr': 6.460208935912814e-08, 'samples': 28594560, 'steps': 148929, 'loss/train': 1.1393177509307861}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:16 - INFO - __main__ - Step 148934: {'lr': 6.412045579043402e-08, 'samples': 28595328, 'steps': 148933, 'loss/train': 1.3165311813354492}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:18 - INFO - __main__ - Step 148938: {'lr': 6.364062409777716e-08, 'samples': 28596096, 'steps': 148937, 'loss/train': 0.7168666124343872}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:20 - INFO - __main__ - Step 148942: {'lr': 6.316259428454374e-08, 'samples': 28596864, 'steps': 148941, 'loss/train': 0.06286223977804184}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:22 - INFO - __main__ - Step 148947: {'lr': 6.256759091627372e-08, 'samples': 28597824, 'steps': 148946, 'loss/train': 1.6736739873886108}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:22 - INFO - __main__ - Step 148947: {'lr': 6.256759091627372e-08, 'samples': 28597824, 'steps': 148946, 'loss/train': 1.6736739873886108}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:22 - INFO - __main__ - Step 148947: {'lr': 6.256759091627372e-08, 'samples': 28597824, 'steps': 148946, 'loss/train': 1.6736739873886108}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:28 - INFO - __main__ - Step 148957: {'lr': 6.13860305326841e-08, 'samples': 28599744, 'steps': 148956, 'loss/train': 1.4744830131530762}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:30 - INFO - __main__ - Step 148962: {'lr': 6.07994735306594e-08, 'samples': 28600704, 'steps': 148961, 'loss/train': 1.447305679321289}}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:30 - INFO - __main__ - Step 148962: {'lr': 6.07994735306594e-08, 'samples': 28600704, 'steps': 148961, 'loss/train': 1.447305679321289}}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:30 - INFO - __main__ - Step 148962: {'lr': 6.07994735306594e-08, 'samples': 28600704, 'steps': 148961, 'loss/train': 1.447305679321289}}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:36 - INFO - __main__ - Step 148973: {'lr': 5.9518958585191854e-08, 'samples': 28602816, 'steps': 148972, 'loss/train': 1.3823879957199097}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:38 - INFO - __main__ - Step 148978: {'lr': 5.894141110601714e-08, 'samples': 28603776, 'steps': 148977, 'loss/train': 1.3801413774490356}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:38 - INFO - __main__ - Step 148978: {'lr': 5.894141110601714e-08, 'samples': 28603776, 'steps': 148977, 'loss/train': 1.3801413774490356}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:41 - INFO - __main__ - Step 148985: {'lr': 5.8137574659167244e-08, 'samples': 28605120, 'steps': 148984, 'loss/train': 1.00044584274292}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:43 - INFO - __main__ - Step 148989: {'lr': 5.7680717184344086e-08, 'samples': 28605888, 'steps': 148988, 'loss/train': 1.1602768898010254}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:46 - INFO - __main__ - Step 148994: {'lr': 5.7112179294660504e-08, 'samples': 28606848, 'steps': 148993, 'loss/train': 1.0971252918243408}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:48 - INFO - __main__ - Step 148998: {'lr': 5.6659376149931216e-08, 'samples': 28607616, 'steps': 148997, 'loss/train': 1.6642168760299683}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:50 - INFO - __main__ - Step 149002: {'lr': 5.620837493500175e-08, 'samples': 28608384, 'steps': 149001, 'loss/train': 1.1181508302688599}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:52 - INFO - __main__ - Step 149006: {'lr': 5.5759175653147254e-08, 'samples': 28609152, 'steps': 149005, 'loss/train': 1.4746838808059692}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:54 - INFO - __main__ - Step 149010: {'lr': 5.531177830755962e-08, 'samples': 28609920, 'steps': 149009, 'loss/train': 1.3261535167694092}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:56 - INFO - __main__ - Step 149014: {'lr': 5.486618290148626e-08, 'samples': 28610688, 'steps': 149013, 'loss/train': 0.6571959853172302}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:11:56 - INFO - __main__ - Step 149014: {'lr': 5.486618290148626e-08, 'samples': 28610688, 'steps': 149013, 'loss/train': 0.6571959853172302}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:00 - INFO - __main__ - Step 149022: {'lr': 5.398039792073317e-08, 'samples': 28612224, 'steps': 149021, 'loss/train': 0.7496612071990967}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:01 - INFO - __main__ - Step 149026: {'lr': 5.354020835240947e-08, 'samples': 28612992, 'steps': 149025, 'loss/train': 1.2927242517471313}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:03 - INFO - __main__ - Step 149030: {'lr': 5.3101820736395353e-08, 'samples': 28613760, 'steps': 149029, 'loss/train': 1.1965105533599854}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:06 - INFO - __main__ - Step 149035: {'lr': 5.255637021656101e-08, 'samples': 28614720, 'steps': 149034, 'loss/train': 1.3804495334625244}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:06 - INFO - __main__ - Step 149035: {'lr': 5.255637021656101e-08, 'samples': 28614720, 'steps': 149034, 'loss/train': 1.3804495334625244}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:10 - INFO - __main__ - Step 149043: {'lr': 5.168950575537568e-08, 'samples': 28616256, 'steps': 149042, 'loss/train': 0.8421028256416321}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:12 - INFO - __main__ - Step 149047: {'lr': 5.1258776471663213e-08, 'samples': 28617024, 'steps': 149046, 'loss/train': 0.8443887829780579}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:14 - INFO - __main__ - Step 149051: {'lr': 5.0829849156691646e-08, 'samples': 28617792, 'steps': 149050, 'loss/train': 1.2289959192276}79}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:16 - INFO - __main__ - Step 149056: {'lr': 5.029622403621148e-08, 'samples': 28618752, 'steps': 149055, 'loss/train': 1.8733577728271484}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:19 - INFO - __main__ - Step 149060: {'lr': 4.987135116216179e-08, 'samples': 28619520, 'steps': 149059, 'loss/train': 1.0438342094421387}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:19 - INFO - __main__ - Step 149060: {'lr': 4.987135116216179e-08, 'samples': 28619520, 'steps': 149059, 'loss/train': 1.0438342094421387}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:22 - INFO - __main__ - Step 149067: {'lr': 4.9132159645776106e-08, 'samples': 28620864, 'steps': 149066, 'loss/train': 0.9060978889465332}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:24 - INFO - __main__ - Step 149072: {'lr': 4.860754442465032e-08, 'samples': 28621824, 'steps': 149071, 'loss/train': 1.4769306182861328}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:24 - INFO - __main__ - Step 149072: {'lr': 4.860754442465032e-08, 'samples': 28621824, 'steps': 149071, 'loss/train': 1.4769306182861328}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:24 - INFO - __main__ - Step 149072: {'lr': 4.860754442465032e-08, 'samples': 28621824, 'steps': 149071, 'loss/train': 1.4769306182861328}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:30 - INFO - __main__ - Step 149083: {'lr': 4.746330187960224e-08, 'samples': 28623936, 'steps': 149082, 'loss/train': 1.165128469467163}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:32 - INFO - __main__ - Step 149088: {'lr': 4.6947696618337976e-08, 'samples': 28624896, 'steps': 149087, 'loss/train': 0.7242974042892456}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:35 - INFO - __main__ - Step 149092: {'lr': 4.6537239658861476e-08, 'samples': 28625664, 'steps': 149091, 'loss/train': 1.083066463470459}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:37 - INFO - __main__ - Step 149096: {'lr': 4.61285847021542e-08, 'samples': 28626432, 'steps': 149095, 'loss/train': 1.287854790687561}9}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:38 - INFO - __main__ - Step 149100: {'lr': 4.572173175113048e-08, 'samples': 28627200, 'steps': 149099, 'loss/train': 1.0389093160629272}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:40 - INFO - __main__ - Step 149104: {'lr': 4.5316680808787926e-08, 'samples': 28627968, 'steps': 149103, 'loss/train': 1.4228005409240723}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:43 - INFO - __main__ - Step 149109: {'lr': 4.4812901209972947e-08, 'samples': 28628928, 'steps': 149108, 'loss/train': 1.3416056632995605}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:43 - INFO - __main__ - Step 149109: {'lr': 4.4812901209972947e-08, 'samples': 28628928, 'steps': 149108, 'loss/train': 1.3416056632995605}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:47 - INFO - __main__ - Step 149116: {'lr': 4.41123400627963e-08, 'samples': 28630272, 'steps': 149115, 'loss/train': 1.0797107219696045}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:48 - INFO - __main__ - Step 149120: {'lr': 4.371449718412745e-08, 'samples': 28631040, 'steps': 149119, 'loss/train': 1.468342661857605}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:50 - INFO - __main__ - Step 149124: {'lr': 4.3318456328572674e-08, 'samples': 28631808, 'steps': 149123, 'loss/train': 1.2808433771133423}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:52 - INFO - __main__ - Step 149128: {'lr': 4.292421749899078e-08, 'samples': 28632576, 'steps': 149127, 'loss/train': 1.0610544681549072}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:52 - INFO - __main__ - Step 149128: {'lr': 4.292421749899078e-08, 'samples': 28632576, 'steps': 149127, 'loss/train': 1.0610544681549072}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:56 - INFO - __main__ - Step 149135: {'lr': 4.223863568081354e-08, 'samples': 28633920, 'steps': 149134, 'loss/train': 1.0247046947479248}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:12:58 - INFO - __main__ - Step 149139: {'lr': 4.184935243728427e-08, 'samples': 28634688, 'steps': 149138, 'loss/train': 1.4658515453338623}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:00 - INFO - __main__ - Step 149144: {'lr': 4.1365282497124455e-08, 'samples': 28635648, 'steps': 149143, 'loss/train': 1.364483118057251}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:00 - INFO - __main__ - Step 149144: {'lr': 4.1365282497124455e-08, 'samples': 28635648, 'steps': 149143, 'loss/train': 1.364483118057251}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:05 - INFO - __main__ - Step 149153: {'lr': 4.050105214123412e-08, 'samples': 28637376, 'steps': 149152, 'loss/train': 1.5539970397949219}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:07 - INFO - __main__ - Step 149157: {'lr': 4.0119878084193816e-08, 'samples': 28638144, 'steps': 149156, 'loss/train': 1.508039116859436}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:08 - INFO - __main__ - Step 149161: {'lr': 3.974050607619129e-08, 'samples': 28638912, 'steps': 149160, 'loss/train': 1.2659348249435425}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:10 - INFO - __main__ - Step 149165: {'lr': 3.936293611994657e-08, 'samples': 28639680, 'steps': 149164, 'loss/train': 1.259766697883606}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:10 - INFO - __main__ - Step 149165: {'lr': 3.936293611994657e-08, 'samples': 28639680, 'steps': 149164, 'loss/train': 1.259766697883606}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:14 - INFO - __main__ - Step 149172: {'lr': 3.8706524891796155e-08, 'samples': 28641024, 'steps': 149171, 'loss/train': 1.6343764066696167}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:16 - INFO - __main__ - Step 149176: {'lr': 3.8333910591908935e-08, 'samples': 28641792, 'steps': 149175, 'loss/train': 1.6841505765914917}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:18 - INFO - __main__ - Step 149181: {'lr': 3.787067686694678e-08, 'samples': 28642752, 'steps': 149180, 'loss/train': 0.8927062749862671}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:21 - INFO - __main__ - Step 149186: {'lr': 3.7410258868941784e-08, 'samples': 28643712, 'steps': 149185, 'loss/train': 1.14633047580719}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:23 - INFO - __main__ - Step 149190: {'lr': 3.7043951797433026e-08, 'samples': 28644480, 'steps': 149189, 'loss/train': 1.3378938436508179}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:23 - INFO - __main__ - Step 149190: {'lr': 3.7043951797433026e-08, 'samples': 28644480, 'steps': 149189, 'loss/train': 1.3378938436508179}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:26 - INFO - __main__ - Step 149197: {'lr': 3.640725065773287e-08, 'samples': 28645824, 'steps': 149196, 'loss/train': 0.6662933826446533}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:28 - INFO - __main__ - Step 149202: {'lr': 3.595584302068522e-08, 'samples': 28646784, 'steps': 149201, 'loss/train': 1.3482815027236938}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:31 - INFO - __main__ - Step 149207: {'lr': 3.55072511321608e-08, 'samples': 28647744, 'steps': 149206, 'loss/train': 1.3746739625930786}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:31 - INFO - __main__ - Step 149207: {'lr': 3.55072511321608e-08, 'samples': 28647744, 'steps': 149206, 'loss/train': 1.3746739625930786}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:35 - INFO - __main__ - Step 149214: {'lr': 3.488395295536684e-08, 'samples': 28649088, 'steps': 149213, 'loss/train': 0.579005777835846}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:36 - INFO - __main__ - Step 149218: {'lr': 3.4530260437093395e-08, 'samples': 28649856, 'steps': 149217, 'loss/train': 1.4993352890014648}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:39 - INFO - __main__ - Step 149223: {'lr': 3.409067897763851e-08, 'samples': 28650816, 'steps': 149222, 'loss/train': 1.2434715032577515}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:41 - INFO - __main__ - Step 149227: {'lr': 3.374104116390941e-08, 'samples': 28651584, 'steps': 149226, 'loss/train': 1.3620963096618652}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:43 - INFO - __main__ - Step 149231: {'lr': 3.3393205445209076e-08, 'samples': 28652352, 'steps': 149230, 'loss/train': 1.4645708799362183}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:43 - INFO - __main__ - Step 149231: {'lr': 3.3393205445209076e-08, 'samples': 28652352, 'steps': 149230, 'loss/train': 1.4645708799362183}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:46 - INFO - __main__ - Step 149238: {'lr': 3.2788829236185534e-08, 'samples': 28653696, 'steps': 149237, 'loss/train': 1.1761510372161865}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:49 - INFO - __main__ - Step 149244: {'lr': 3.227518510842686e-08, 'samples': 28654848, 'steps': 149243, 'loss/train': 0.9963176250457764}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:49 - INFO - __main__ - Step 149244: {'lr': 3.227518510842686e-08, 'samples': 28654848, 'steps': 149243, 'loss/train': 0.9963176250457764}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:53 - INFO - __main__ - Step 149251: {'lr': 3.168105836440227e-08, 'samples': 28656192, 'steps': 149250, 'loss/train': 1.2969226837158203}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:54 - INFO - __main__ - Step 149255: {'lr': 3.134403526805496e-08, 'samples': 28656960, 'steps': 149254, 'loss/train': 1.5305206775665283}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:56 - INFO - __main__ - Step 149259: {'lr': 3.100881428400038e-08, 'samples': 28657728, 'steps': 149258, 'loss/train': 1.1719932556152344}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:59 - INFO - __main__ - Step 149265: {'lr': 3.050936177378949e-08, 'samples': 28658880, 'steps': 149264, 'loss/train': 1.3028775453567505}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:13:59 - INFO - __main__ - Step 149265: {'lr': 3.050936177378949e-08, 'samples': 28658880, 'steps': 149264, 'loss/train': 1.3028775453567505}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:02 - INFO - __main__ - Step 149271: {'lr': 3.0013964029795835e-08, 'samples': 28660032, 'steps': 149270, 'loss/train': 1.3969788551330566}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:04 - INFO - __main__ - Step 149275: {'lr': 2.968595151903175e-08, 'samples': 28660800, 'steps': 149274, 'loss/train': 1.0537910461425781}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:04 - INFO - __main__ - Step 149275: {'lr': 2.968595151903175e-08, 'samples': 28660800, 'steps': 149274, 'loss/train': 1.0537910461425781}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:09 - INFO - __main__ - Step 149284: {'lr': 2.8954512390155783e-08, 'samples': 28662528, 'steps': 149283, 'loss/train': 1.2382670640945435}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:11 - INFO - __main__ - Step 149288: {'lr': 2.8632356791791214e-08, 'samples': 28663296, 'steps': 149287, 'loss/train': 1.0901999473571777}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:12 - INFO - __main__ - Step 149292: {'lr': 2.831200332528705e-08, 'samples': 28664064, 'steps': 149291, 'loss/train': 1.2373827695846558}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:14 - INFO - __main__ - Step 149296: {'lr': 2.799345199291925e-08, 'samples': 28664832, 'steps': 149295, 'loss/train': 1.077060580253601}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:17 - INFO - __main__ - Step 149301: {'lr': 2.7597797082168365e-08, 'samples': 28665792, 'steps': 149300, 'loss/train': 1.7098400592803955}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:17 - INFO - __main__ - Step 149301: {'lr': 2.7597797082168365e-08, 'samples': 28665792, 'steps': 149300, 'loss/train': 1.7098400592803955}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:20 - INFO - __main__ - Step 149308: {'lr': 2.7048610823782182e-08, 'samples': 28667136, 'steps': 149307, 'loss/train': 1.0865931510925293}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:22 - INFO - __main__ - Step 149312: {'lr': 2.67372680509137e-08, 'samples': 28667904, 'steps': 149311, 'loss/train': 1.3286848068237305}3}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:25 - INFO - __main__ - Step 149317: {'lr': 2.6350623852228195e-08, 'samples': 28668864, 'steps': 149316, 'loss/train': 0.8327307105064392}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:25 - INFO - __main__ - Step 149317: {'lr': 2.6350623852228195e-08, 'samples': 28668864, 'steps': 149316, 'loss/train': 0.8327307105064392}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:28 - INFO - __main__ - Step 149324: {'lr': 2.5814052614370197e-08, 'samples': 28670208, 'steps': 149323, 'loss/train': 1.3125114440917969}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:31 - INFO - __main__ - Step 149328: {'lr': 2.5509918436916745e-08, 'samples': 28670976, 'steps': 149327, 'loss/train': 1.7433141469955444}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:33 - INFO - __main__ - Step 149332: {'lr': 2.520758641383347e-08, 'samples': 28671744, 'steps': 149331, 'loss/train': 1.0757824182510376}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:35 - INFO - __main__ - Step 149336: {'lr': 2.4907056547285312e-08, 'samples': 28672512, 'steps': 149335, 'loss/train': 1.337040901184082}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:36 - INFO - __main__ - Step 149340: {'lr': 2.46083288394372e-08, 'samples': 28673280, 'steps': 149339, 'loss/train': 1.451332688331604}2}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:39 - INFO - __main__ - Step 149345: {'lr': 2.4237453493641815e-08, 'samples': 28674240, 'steps': 149344, 'loss/train': 1.2851461172103882}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:41 - INFO - __main__ - Step 149349: {'lr': 2.3942780650765273e-08, 'samples': 28675008, 'steps': 149348, 'loss/train': 1.5162914991378784}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:43 - INFO - __main__ - Step 149353: {'lr': 2.3649909973555427e-08, 'samples': 28675776, 'steps': 149352, 'loss/train': 1.4088304042816162}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:45 - INFO - __main__ - Step 149357: {'lr': 2.335884146409395e-08, 'samples': 28676544, 'steps': 149356, 'loss/train': 1.2092705965042114}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:46 - INFO - __main__ - Step 149361: {'lr': 2.3069575124545773e-08, 'samples': 28677312, 'steps': 149360, 'loss/train': 1.2438743114471436}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:49 - INFO - __main__ - Step 149365: {'lr': 2.278211095693705e-08, 'samples': 28678080, 'steps': 149364, 'loss/train': 0.8335750699043274}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:51 - INFO - __main__ - Step 149370: {'lr': 2.2425315054891426e-08, 'samples': 28679040, 'steps': 149369, 'loss/train': 1.436560034751892}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:53 - INFO - __main__ - Step 149374: {'lr': 2.2141905781758454e-08, 'samples': 28679808, 'steps': 149373, 'loss/train': 1.3251935243606567}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:53 - INFO - __main__ - Step 149374: {'lr': 2.2141905781758454e-08, 'samples': 28679808, 'steps': 149373, 'loss/train': 1.3251935243606567}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:57 - INFO - __main__ - Step 149381: {'lr': 2.165027604736358e-08, 'samples': 28681152, 'steps': 149380, 'loss/train': 1.0623489618301392}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:14:59 - INFO - __main__ - Step 149385: {'lr': 2.1371822770371864e-08, 'samples': 28681920, 'steps': 149384, 'loss/train': 1.4446430206298828}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:01 - INFO - __main__ - Step 149389: {'lr': 2.1095171677587566e-08, 'samples': 28682688, 'steps': 149388, 'loss/train': 1.3916548490524292}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:03 - INFO - __main__ - Step 149393: {'lr': 2.0820322770981337e-08, 'samples': 28683456, 'steps': 149392, 'loss/train': 1.1905814409255981}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:05 - INFO - __main__ - Step 149397: {'lr': 2.0547276052579333e-08, 'samples': 28684224, 'steps': 149396, 'loss/train': 1.8221162557601929}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:07 - INFO - __main__ - Step 149402: {'lr': 2.0208501984669834e-08, 'samples': 28685184, 'steps': 149401, 'loss/train': 1.1539918184280396}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:09 - INFO - __main__ - Step 149406: {'lr': 1.993951019679807e-08, 'samples': 28685952, 'steps': 149405, 'loss/train': 1.0088876485824585}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:09 - INFO - __main__ - Step 149406: {'lr': 1.993951019679807e-08, 'samples': 28685952, 'steps': 149405, 'loss/train': 1.0088876485824585}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:12 - INFO - __main__ - Step 149413: {'lr': 1.947311109978811e-08, 'samples': 28687296, 'steps': 149412, 'loss/train': 1.2453014850616455}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:12 - INFO - __main__ - Step 149413: {'lr': 1.947311109978811e-08, 'samples': 28687296, 'steps': 149412, 'loss/train': 1.2453014850616455}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:17 - INFO - __main__ - Step 149421: {'lr': 1.8946841803035852e-08, 'samples': 28688832, 'steps': 149420, 'loss/train': 1.422516942024231}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:19 - INFO - __main__ - Step 149425: {'lr': 1.8686410456214554e-08, 'samples': 28689600, 'steps': 149424, 'loss/train': 0.9834445714950562}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:20 - INFO - __main__ - Step 149429: {'lr': 1.842778131297407e-08, 'samples': 28690368, 'steps': 149428, 'loss/train': 1.1854989528656006}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:23 - INFO - __main__ - Step 149433: {'lr': 1.817095437511851e-08, 'samples': 28691136, 'steps': 149432, 'loss/train': 1.9547779560089111}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:25 - INFO - __main__ - Step 149438: {'lr': 1.7852455057032478e-08, 'samples': 28692096, 'steps': 149437, 'loss/train': 1.5269670486450195}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:25 - INFO - __main__ - Step 149438: {'lr': 1.7852455057032478e-08, 'samples': 28692096, 'steps': 149437, 'loss/train': 1.5269670486450195}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:28 - INFO - __main__ - Step 149445: {'lr': 1.741128681262416e-08, 'samples': 28693440, 'steps': 149444, 'loss/train': 1.005128264427185}5}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:30 - INFO - __main__ - Step 149449: {'lr': 1.7161668714876787e-08, 'samples': 28694208, 'steps': 149448, 'loss/train': 1.4719164371490479}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:33 - INFO - __main__ - Step 149454: {'lr': 1.685218045710535e-08, 'samples': 28695168, 'steps': 149453, 'loss/train': 1.5073233842849731}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:33 - INFO - __main__ - Step 149454: {'lr': 1.685218045710535e-08, 'samples': 28695168, 'steps': 149453, 'loss/train': 1.5073233842849731}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:37 - INFO - __main__ - Step 149462: {'lr': 1.6362856450652385e-08, 'samples': 28696704, 'steps': 149461, 'loss/train': 1.2535442113876343}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:37 - INFO - __main__ - Step 149462: {'lr': 1.6362856450652385e-08, 'samples': 28696704, 'steps': 149461, 'loss/train': 1.2535442113876343}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:40 - INFO - __main__ - Step 149469: {'lr': 1.5940611479914325e-08, 'samples': 28698048, 'steps': 149468, 'loss/train': 1.565960168838501}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:43 - INFO - __main__ - Step 149474: {'lr': 1.5642387097614118e-08, 'samples': 28699008, 'steps': 149473, 'loss/train': 0.9414656758308411}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:45 - INFO - __main__ - Step 149479: {'lr': 1.5346978692681867e-08, 'samples': 28699968, 'steps': 149478, 'loss/train': 1.1214573383331299}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:45 - INFO - __main__ - Step 149479: {'lr': 1.5346978692681867e-08, 'samples': 28699968, 'steps': 149478, 'loss/train': 1.1214573383331299}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:49 - INFO - __main__ - Step 149486: {'lr': 1.49381377740343e-08, 'samples': 28701312, 'steps': 149485, 'loss/train': 1.229589819908142}99}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:51 - INFO - __main__ - Step 149490: {'lr': 1.4706992458424218e-08, 'samples': 28702080, 'steps': 149489, 'loss/train': 0.8421230316162109}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:53 - INFO - __main__ - Step 149495: {'lr': 1.4420595203207264e-08, 'samples': 28703040, 'steps': 149494, 'loss/train': 1.3662739992141724}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:55 - INFO - __main__ - Step 149499: {'lr': 1.419350491249638e-08, 'samples': 28703808, 'steps': 149498, 'loss/train': 1.772628903388977}4}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:57 - INFO - __main__ - Step 149503: {'lr': 1.396821685772931e-08, 'samples': 28704576, 'steps': 149502, 'loss/train': 1.5067238807678223}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:15:59 - INFO - __main__ - Step 149507: {'lr': 1.3744731040515879e-08, 'samples': 28705344, 'steps': 149506, 'loss/train': 1.2887166738510132}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:01 - INFO - __main__ - Step 149511: {'lr': 1.3523047462493666e-08, 'samples': 28706112, 'steps': 149510, 'loss/train': 1.4277353286743164}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:03 - INFO - __main__ - Step 149516: {'lr': 1.3248477391258895e-08, 'samples': 28707072, 'steps': 149515, 'loss/train': 1.4058796167373657}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:05 - INFO - __main__ - Step 149520: {'lr': 1.3030848857231714e-08, 'samples': 28707840, 'steps': 149519, 'loss/train': 1.1614617109298706}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:07 - INFO - __main__ - Step 149524: {'lr': 1.2815022567530532e-08, 'samples': 28708608, 'steps': 149523, 'loss/train': 1.4388784170150757}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:09 - INFO - __main__ - Step 149528: {'lr': 1.2600998523709661e-08, 'samples': 28709376, 'steps': 149527, 'loss/train': 0.6732990145683289}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:11 - INFO - __main__ - Step 149532: {'lr': 1.2388776727323414e-08, 'samples': 28710144, 'steps': 149531, 'loss/train': 1.0698994398117065}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:11 - INFO - __main__ - Step 149532: {'lr': 1.2388776727323414e-08, 'samples': 28710144, 'steps': 149531, 'loss/train': 1.0698994398117065}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:15 - INFO - __main__ - Step 149538: {'lr': 1.207382325002504e-08, 'samples': 28711296, 'steps': 149537, 'loss/train': 0.5636351704597473}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:17 - INFO - __main__ - Step 149542: {'lr': 1.1866107078900568e-08, 'samples': 28712064, 'steps': 149541, 'loss/train': 1.3959885835647583}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:19 - INFO - __main__ - Step 149548: {'lr': 1.1557912046517016e-08, 'samples': 28713216, 'steps': 149547, 'loss/train': 1.4081352949142456}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:22 - INFO - __main__ - Step 149552: {'lr': 1.1354701509980193e-08, 'samples': 28713984, 'steps': 149551, 'loss/train': 0.9747281670570374}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:24 - INFO - __main__ - Step 149556: {'lr': 1.1153293229843042e-08, 'samples': 28714752, 'steps': 149555, 'loss/train': 1.5912730693817139}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:24 - INFO - __main__ - Step 149556: {'lr': 1.1153293229843042e-08, 'samples': 28714752, 'steps': 149555, 'loss/train': 1.5912730693817139}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:27 - INFO - __main__ - Step 149563: {'lr': 1.0805165423483088e-08, 'samples': 28716096, 'steps': 149562, 'loss/train': 1.2670412063598633}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:29 - INFO - __main__ - Step 149568: {'lr': 1.0559881942401317e-08, 'samples': 28717056, 'steps': 149567, 'loss/train': 1.588313341140747}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:31 - INFO - __main__ - Step 149572: {'lr': 1.0365682702351276e-08, 'samples': 28717824, 'steps': 149571, 'loss/train': 0.8436179757118225}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:34 - INFO - __main__ - Step 149576: {'lr': 1.0173285725806336e-08, 'samples': 28718592, 'steps': 149575, 'loss/train': 1.1963571310043335}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:35 - INFO - __main__ - Step 149580: {'lr': 9.982691014209788e-09, 'samples': 28719360, 'steps': 149579, 'loss/train': 0.08100341260433197}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:37 - INFO - __main__ - Step 149584: {'lr': 9.793898568921655e-09, 'samples': 28720128, 'steps': 149583, 'loss/train': 0.6572380065917969}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:39 - INFO - __main__ - Step 149589: {'lr': 9.560442451372885e-09, 'samples': 28721088, 'steps': 149588, 'loss/train': 0.8279416561126709}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:42 - INFO - __main__ - Step 149594: {'lr': 9.329802379692831e-09, 'samples': 28722048, 'steps': 149593, 'loss/train': 1.4981257915496826}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:42 - INFO - __main__ - Step 149594: {'lr': 9.329802379692831e-09, 'samples': 28722048, 'steps': 149593, 'loss/train': 1.4981257915496826}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:45 - INFO - __main__ - Step 149601: {'lr': 9.011637241318482e-09, 'samples': 28723392, 'steps': 149600, 'loss/train': 1.6478341817855835}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:47 - INFO - __main__ - Step 149605: {'lr': 8.832306716166637e-09, 'samples': 28724160, 'steps': 149604, 'loss/train': 0.8971133828163147}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:50 - INFO - __main__ - Step 149610: {'lr': 8.610678008530437e-09, 'samples': 28725120, 'steps': 149609, 'loss/train': 0.6303962469100952}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:52 - INFO - __main__ - Step 149614: {'lr': 8.43540260300757e-09, 'samples': 28725888, 'steps': 149613, 'loss/train': 0.8345909714698792}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:54 - INFO - __main__ - Step 149618: {'lr': 8.261929474978614e-09, 'samples': 28726656, 'steps': 149617, 'loss/train': 1.3995612859725952}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:54 - INFO - __main__ - Step 149618: {'lr': 8.261929474978614e-09, 'samples': 28726656, 'steps': 149617, 'loss/train': 1.3995612859725952}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:16:57 - INFO - __main__ - Step 149625: {'lr': 7.962688234802729e-09, 'samples': 28728000, 'steps': 149624, 'loss/train': 1.2762154340744019}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:00 - INFO - __main__ - Step 149630: {'lr': 7.752323768039914e-09, 'samples': 28728960, 'steps': 149629, 'loss/train': 0.0524417944252491}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:02 - INFO - __main__ - Step 149634: {'lr': 7.586059762226816e-09, 'samples': 28729728, 'steps': 149633, 'loss/train': 1.7539290189743042}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:04 - INFO - __main__ - Step 149638: {'lr': 7.4215980399861e-09, 'samples': 28730496, 'steps': 149637, 'loss/train': 1.5418555736541748}2}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:05 - INFO - __main__ - Step 149642: {'lr': 7.258938602511256e-09, 'samples': 28731264, 'steps': 149641, 'loss/train': 1.196815013885498}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:07 - INFO - __main__ - Step 149646: {'lr': 7.0980814509957745e-09, 'samples': 28732032, 'steps': 149645, 'loss/train': 1.4907102584838867}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:07 - INFO - __main__ - Step 149646: {'lr': 7.0980814509957745e-09, 'samples': 28732032, 'steps': 149645, 'loss/train': 1.4907102584838867}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:07 - INFO - __main__ - Step 149646: {'lr': 7.0980814509957745e-09, 'samples': 28732032, 'steps': 149645, 'loss/train': 1.4907102584838867}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:13 - INFO - __main__ - Step 149657: {'lr': 6.665017330625966e-09, 'samples': 28734144, 'steps': 149656, 'loss/train': 1.0507755279541016}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:15 - INFO - __main__ - Step 149661: {'lr': 6.510918761609697e-09, 'samples': 28734912, 'steps': 149660, 'loss/train': 0.7859552502632141}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:15 - INFO - __main__ - Step 149661: {'lr': 6.510918761609697e-09, 'samples': 28734912, 'steps': 149660, 'loss/train': 0.7859552502632141}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:18 - INFO - __main__ - Step 149666: {'lr': 6.320830022665813e-09, 'samples': 28735872, 'steps': 149665, 'loss/train': 1.0021696090698242}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:21 - INFO - __main__ - Step 149672: {'lr': 6.096440764824518e-09, 'samples': 28737024, 'steps': 149671, 'loss/train': 1.678890585899353}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:23 - INFO - __main__ - Step 149677: {'lr': 5.912547409808733e-09, 'samples': 28737984, 'steps': 149676, 'loss/train': 1.6837669610977173}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:23 - INFO - __main__ - Step 149677: {'lr': 5.912547409808733e-09, 'samples': 28737984, 'steps': 149676, 'loss/train': 1.6837669610977173}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:28 - INFO - __main__ - Step 149685: {'lr': 5.624175503537998e-09, 'samples': 28739520, 'steps': 149684, 'loss/train': 1.2617255449295044}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:29 - INFO - __main__ - Step 149689: {'lr': 5.482692996466865e-09, 'samples': 28740288, 'steps': 149688, 'loss/train': 1.0971020460128784}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:32 - INFO - __main__ - Step 149693: {'lr': 5.343012788150415e-09, 'samples': 28741056, 'steps': 149692, 'loss/train': 1.4664844274520874}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:32 - INFO - __main__ - Step 149693: {'lr': 5.343012788150415e-09, 'samples': 28741056, 'steps': 149692, 'loss/train': 1.4664844274520874}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:32 - INFO - __main__ - Step 149693: {'lr': 5.343012788150415e-09, 'samples': 28741056, 'steps': 149692, 'loss/train': 1.4664844274520874}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:38 - INFO - __main__ - Step 149704: {'lr': 4.968185326387209e-09, 'samples': 28743168, 'steps': 149703, 'loss/train': 1.1823065280914307}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:38 - INFO - __main__ - Step 149704: {'lr': 4.968185326387209e-09, 'samples': 28743168, 'steps': 149703, 'loss/train': 1.1823065280914307}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:42 - INFO - __main__ - Step 149713: {'lr': 4.671646262610629e-09, 'samples': 28744896, 'steps': 149712, 'loss/train': 1.4903628826141357}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:44 - INFO - __main__ - Step 149717: {'lr': 4.542779867472424e-09, 'samples': 28745664, 'steps': 149716, 'loss/train': 1.2481846809387207}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:46 - INFO - __main__ - Step 149721: {'lr': 4.415715777861262e-09, 'samples': 28746432, 'steps': 149720, 'loss/train': 1.2117501497268677}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:48 - INFO - __main__ - Step 149726: {'lr': 4.2594201594137184e-09, 'samples': 28747392, 'steps': 149725, 'loss/train': 1.3866337537765503}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:48 - INFO - __main__ - Step 149726: {'lr': 4.2594201594137184e-09, 'samples': 28747392, 'steps': 149725, 'loss/train': 1.3866337537765503}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:51 - INFO - __main__ - Step 149732: {'lr': 4.0755826767602256e-09, 'samples': 28748544, 'steps': 149731, 'loss/train': 1.3939422369003296}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:54 - INFO - __main__ - Step 149737: {'lr': 3.9254824928114385e-09, 'samples': 28749504, 'steps': 149736, 'loss/train': 1.3015015125274658}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:54 - INFO - __main__ - Step 149737: {'lr': 3.9254824928114385e-09, 'samples': 28749504, 'steps': 149736, 'loss/train': 1.3015015125274658}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:58 - INFO - __main__ - Step 149745: {'lr': 3.69117970663857e-09, 'samples': 28751040, 'steps': 149744, 'loss/train': 1.423858404159546}58}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:17:59 - INFO - __main__ - Step 149749: {'lr': 3.576731780641218e-09, 'samples': 28751808, 'steps': 149748, 'loss/train': 1.3854159116744995}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:01 - INFO - __main__ - Step 149753: {'lr': 3.4640861671098035e-09, 'samples': 28752576, 'steps': 149752, 'loss/train': 1.5855780839920044}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:04 - INFO - __main__ - Step 149758: {'lr': 3.3258136533931284e-09, 'samples': 28753536, 'steps': 149757, 'loss/train': 1.6739147901535034}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:06 - INFO - __main__ - Step 149762: {'lr': 3.217223245921552e-09, 'samples': 28754304, 'steps': 149761, 'loss/train': 1.6387925148010254}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:06 - INFO - __main__ - Step 149762: {'lr': 3.217223245921552e-09, 'samples': 28754304, 'steps': 149761, 'loss/train': 1.6387925148010254}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:09 - INFO - __main__ - Step 149769: {'lr': 3.0315268539637953e-09, 'samples': 28755648, 'steps': 149768, 'loss/train': 1.6297430992126465}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:12 - INFO - __main__ - Step 149774: {'lr': 2.9022659170652167e-09, 'samples': 28756608, 'steps': 149773, 'loss/train': 1.4155726432800293}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:12 - INFO - __main__ - Step 149774: {'lr': 2.9022659170652167e-09, 'samples': 28756608, 'steps': 149773, 'loss/train': 1.4155726432800293}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:16 - INFO - __main__ - Step 149782: {'lr': 2.701305949981414e-09, 'samples': 28758144, 'steps': 149781, 'loss/train': 1.3061957359313965}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:18 - INFO - __main__ - Step 149786: {'lr': 2.6035294442838807e-09, 'samples': 28758912, 'steps': 149785, 'loss/train': 0.8828412890434265}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:20 - INFO - __main__ - Step 149790: {'lr': 2.507555258102201e-09, 'samples': 28759680, 'steps': 149789, 'loss/train': 1.4516167640686035}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:22 - INFO - __main__ - Step 149794: {'lr': 2.4133833921025084e-09, 'samples': 28760448, 'steps': 149793, 'loss/train': 1.5574758052825928}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:24 - INFO - __main__ - Step 149799: {'lr': 2.2982030734031158e-09, 'samples': 28761408, 'steps': 149798, 'loss/train': 1.4518661499023438}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:26 - INFO - __main__ - Step 149803: {'lr': 2.2080864303386515e-09, 'samples': 28762176, 'steps': 149802, 'loss/train': 1.5713214874267578}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:28 - INFO - __main__ - Step 149807: {'lr': 2.1197721095933544e-09, 'samples': 28762944, 'steps': 149806, 'loss/train': 1.6265714168548584}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:30 - INFO - __main__ - Step 149811: {'lr': 2.033260111861113e-09, 'samples': 28763712, 'steps': 149810, 'loss/train': 1.7272670269012451}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:32 - INFO - __main__ - Step 149816: {'lr': 1.927654632360998e-09, 'samples': 28764672, 'steps': 149815, 'loss/train': 1.662630319595337}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:34 - INFO - __main__ - Step 149820: {'lr': 1.8451978636147005e-09, 'samples': 28765440, 'steps': 149819, 'loss/train': 0.808184802532196}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:36 - INFO - __main__ - Step 149824: {'lr': 1.7645434198243494e-09, 'samples': 28766208, 'steps': 149823, 'loss/train': 1.4029489755630493}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:36 - INFO - __main__ - Step 149824: {'lr': 1.7645434198243494e-09, 'samples': 28766208, 'steps': 149823, 'loss/train': 1.4029489755630493}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:40 - INFO - __main__ - Step 149831: {'lr': 1.6277349894344707e-09, 'samples': 28767552, 'steps': 149830, 'loss/train': 2.1862733364105225}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:40 - INFO - __main__ - Step 149831: {'lr': 1.6277349894344707e-09, 'samples': 28767552, 'steps': 149830, 'loss/train': 2.1862733364105225}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:44 - INFO - __main__ - Step 149839: {'lr': 1.4781412222553136e-09, 'samples': 28769088, 'steps': 149838, 'loss/train': 1.4478543996810913}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:46 - INFO - __main__ - Step 149843: {'lr': 1.4060478298327794e-09, 'samples': 28769856, 'steps': 149842, 'loss/train': 1.3922572135925293}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:48 - INFO - __main__ - Step 149847: {'lr': 1.3357567655303271e-09, 'samples': 28770624, 'steps': 149846, 'loss/train': 0.43890175223350525}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:50 - INFO - __main__ - Step 149852: {'lr': 1.2504274598812426e-09, 'samples': 28771584, 'steps': 149851, 'loss/train': 1.3307799100875854}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:52 - INFO - __main__ - Step 149856: {'lr': 1.1841916357224758e-09, 'samples': 28772352, 'steps': 149855, 'loss/train': 1.1148765087127686}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:52 - INFO - __main__ - Step 149856: {'lr': 1.1841916357224758e-09, 'samples': 28772352, 'steps': 149855, 'loss/train': 1.1148765087127686}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:55 - INFO - __main__ - Step 149863: {'lr': 1.0726157996798768e-09, 'samples': 28773696, 'steps': 149862, 'loss/train': 1.4477200508117676}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:18:57 - INFO - __main__ - Step 149867: {'lr': 1.011336383416328e-09, 'samples': 28774464, 'steps': 149866, 'loss/train': 1.147086262702942}6}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:00 - INFO - __main__ - Step 149872: {'lr': 9.372716411049087e-10, 'samples': 28775424, 'steps': 149871, 'loss/train': 1.169695258140564}6}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:00 - INFO - __main__ - Step 149872: {'lr': 9.372716411049087e-10, 'samples': 28775424, 'steps': 149871, 'loss/train': 1.169695258140564}6}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:04 - INFO - __main__ - Step 149880: {'lr': 8.246256311406076e-10, 'samples': 28776960, 'steps': 149879, 'loss/train': 1.4123514890670776}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:04 - INFO - __main__ - Step 149880: {'lr': 8.246256311406076e-10, 'samples': 28776960, 'steps': 149879, 'loss/train': 1.4123514890670776}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:07 - INFO - __main__ - Step 149887: {'lr': 7.319742752753822e-10, 'samples': 28778304, 'steps': 149886, 'loss/train': 1.9748057126998901}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:10 - INFO - __main__ - Step 149892: {'lr': 6.691741096243043e-10, 'samples': 28779264, 'steps': 149891, 'loss/train': 1.4535033702850342}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:12 - INFO - __main__ - Step 149896: {'lr': 6.209616021934927e-10, 'samples': 28780032, 'steps': 149895, 'loss/train': 1.4988442659378052}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:14 - INFO - __main__ - Step 149900: {'lr': 5.745514285726561e-10, 'samples': 28780800, 'steps': 149899, 'loss/train': 1.6253581047058105}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:16 - INFO - __main__ - Step 149905: {'lr': 5.19073243943069e-10, 'samples': 28781760, 'steps': 149904, 'loss/train': 1.4319241046905518}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:18 - INFO - __main__ - Step 149909: {'lr': 4.767183225673488e-10, 'samples': 28782528, 'steps': 149908, 'loss/train': 1.309288740158081}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:18 - INFO - __main__ - Step 149909: {'lr': 4.767183225673488e-10, 'samples': 28782528, 'steps': 149908, 'loss/train': 1.309288740158081}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:21 - INFO - __main__ - Step 149916: {'lr': 4.069340785450848e-10, 'samples': 28783872, 'steps': 149915, 'loss/train': 1.3984349966049194}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:23 - INFO - __main__ - Step 149920: {'lr': 3.695355786648147e-10, 'samples': 28784640, 'steps': 149919, 'loss/train': 1.0588749647140503}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:26 - INFO - __main__ - Step 149925: {'lr': 3.253219883203595e-10, 'samples': 28785600, 'steps': 149924, 'loss/train': 0.9910423159599304}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:28 - INFO - __main__ - Step 149930: {'lr': 2.8392454784476053e-10, 'samples': 28786560, 'steps': 149929, 'loss/train': 1.2449990510940552}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:28 - INFO - __main__ - Step 149930: {'lr': 2.8392454784476053e-10, 'samples': 28786560, 'steps': 149929, 'loss/train': 1.2449990510940552}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:32 - INFO - __main__ - Step 149937: {'lr': 2.3069926388674667e-10, 'samples': 28787904, 'steps': 149936, 'loss/train': 1.7680420875549316}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:34 - INFO - __main__ - Step 149941: {'lr': 2.0276302867272557e-10, 'samples': 28788672, 'steps': 149940, 'loss/train': 1.400766372680664}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:36 - INFO - __main__ - Step 149946: {'lr': 1.7037727081947728e-10, 'samples': 28789632, 'steps': 149945, 'loss/train': 0.8468477725982666}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:38 - INFO - __main__ - Step 149951: {'lr': 1.40807664611442e-10, 'samples': 28790592, 'steps': 149950, 'loss/train': 1.3526259660720825}6}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:38 - INFO - __main__ - Step 149951: {'lr': 1.40807664611442e-10, 'samples': 28790592, 'steps': 149950, 'loss/train': 1.3526259660720825}6}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:42 - INFO - __main__ - Step 149958: {'lr': 1.0414135129255265e-10, 'samples': 28791936, 'steps': 149957, 'loss/train': 0.9350542426109314}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:44 - INFO - __main__ - Step 149962: {'lr': 8.566738629589565e-11, 'samples': 28792704, 'steps': 149961, 'loss/train': 1.1058440208435059}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:46 - INFO - __main__ - Step 149966: {'lr': 6.899575896723852e-11, 'samples': 28793472, 'steps': 149965, 'loss/train': 1.4233659505844116}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:48 - INFO - __main__ - Step 149971: {'lr': 5.069076230990177e-11, 'samples': 28794432, 'steps': 149970, 'loss/train': 1.349865198135376}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:48 - INFO - __main__ - Step 149971: {'lr': 5.069076230990177e-11, 'samples': 28794432, 'steps': 149970, 'loss/train': 1.349865198135376}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:52 - INFO - __main__ - Step 149978: {'lr': 2.979490404397822e-11, 'samples': 28795776, 'steps': 149977, 'loss/train': 1.3207296133041382}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:54 - INFO - __main__ - Step 149982: {'lr': 2.0332628414276854e-11, 'samples': 28796544, 'steps': 149981, 'loss/train': 1.255293607711792}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:56 - INFO - __main__ - Step 149986: {'lr': 1.2672690896664563e-11, 'samples': 28797312, 'steps': 149985, 'loss/train': 1.0476105213165283}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:19:58 - INFO - __main__ - Step 149990: {'lr': 6.815091574408072e-12, 'samples': 28798080, 'steps': 149989, 'loss/train': 0.9487575888633728}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:20:00 - INFO - __main__ - Step 149994: {'lr': 2.759830475262959e-12, 'samples': 28798848, 'steps': 149993, 'loss/train': 1.2347151041030884}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:20:02 - INFO - __main__ - Step 149998: {'lr': 5.069076269847983e-13, 'samples': 28799616, 'steps': 149997, 'loss/train': 1.498703122138977}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:20:02 - INFO - __main__ - Step 149998: {'lr': 5.069076269847983e-13, 'samples': 28799616, 'steps': 149997, 'loss/train': 1.498703122138977}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:20:02 - INFO - __main__ - Step 149998: {'lr': 5.069076269847983e-13, 'samples': 28799616, 'steps': 149997, 'loss/train': 1.498703122138977}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+11/07/2021 18:20:02 - INFO - __main__ - Step 149998: {'lr': 5.069076269847983e-13, 'samples': 28799616, 'steps': 149997, 'loss/train': 1.498703122138977}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+	- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
+11/07/2021 18:20:02 - INFO - __main__ - Step 149998: {'lr': 5.069076269847983e-13, 'samples': 28799616, 'steps': 149997, 'loss/train': 1.498703122138977}}}}}ils.clip_grad_norm_; continuing anyway. Note that the default behavior will change in a future release to error out if a non-finite total norm is encountered. At that point, setting error_if_nonfinite=false will be required to retain the old behavior.
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+	- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+	- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+	- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+	- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+	- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible
+	- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
+huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
+To disable this warning, you can either:
+	- Avoid using `tokenizers` before the fork if possible