zuazo/whisper-large-v2-pt

Whisper Large-V2 Portuguese

This model is a fine-tuned version of openai/whisper-large-v2 on the mozilla-foundation/common_voice_13_0 pt dataset. It achieves the following results on the evaluation set:

• Loss: 0.4115
• Wer: 6.4502

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0424	3.53	1000	0.1879	5.9886
0.0093	7.05	2000	0.2493	6.2744
0.0049	10.58	3000	0.2656	6.3500
0.0031	14.11	4000	0.2850	6.4256
0.0029	17.64	5000	0.3013	6.5833
0.0053	21.16	6000	0.2810	6.7969
0.0033	24.69	7000	0.3072	6.8905
0.0007	28.22	8000	0.3210	6.7312
0.0021	31.75	9000	0.3311	6.9924
0.0006	35.27	10000	0.3188	6.7295
0.0002	38.8	11000	0.3336	6.6589
0.0004	42.33	12000	0.3465	6.9086
0.0004	45.86	13000	0.3340	6.9924
0.0001	49.38	14000	0.3607	6.8199
0.0001	52.91	15000	0.3779	6.6112
0.0	56.44	16000	0.3884	6.5505
0.0	59.96	17000	0.3966	6.4897
0.0	63.49	18000	0.4039	6.4650
0.0	67.02	19000	0.4091	6.4486
0.0	70.55	20000	0.4115	6.4502

Framework versions

• Transformers 4.37.2
• Pytorch 2.0.1+cu117
• Datasets 2.14.4
• Tokenizers 0.15.1