DandinPower/breeze_7b_lora_full_text

breeze_7b_lora

This model is a fine-tuned version of MediaTek-Research/Breeze-7B-Instruct-v1_0 on the DandinPower/ZH-Reading-Comprehension-Breeze-Instruct dataset. It achieves the following results on the evaluation set:

• Loss: 0.9671

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: 0.0001
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• total_eval_batch_size: 2
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 700
• num_epochs: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss
2.2919	0.3690	250	2.2932
2.2105	0.7380	500	2.1866
1.9287	1.1070	750	1.9796
1.8181	1.4760	1000	1.8416
1.6765	1.8450	1250	1.7156
1.4271	2.2140	1500	1.6054
1.3595	2.5830	1750	1.5071
1.2794	2.9520	2000	1.4263
1.0636	3.3210	2250	1.3707
1.0272	3.6900	2500	1.3044
0.8977	4.0590	2750	1.2597
0.8923	4.4280	3000	1.2184
0.8628	4.7970	3250	1.1737
0.6994	5.1661	3500	1.1514
0.7201	5.5351	3750	1.1209
0.7237	5.9041	4000	1.0931
0.6468	6.2731	4250	1.0740
0.6052	6.6421	4500	1.0472
0.5737	7.0111	4750	1.0360
0.5419	7.3801	5000	1.0246
0.5539	7.7491	5250	1.0027
0.4615	8.1181	5500	0.9947
0.4782	8.4871	5750	0.9851
0.4809	8.8561	6000	0.9699
0.4284	9.2251	6250	0.9738
0.4332	9.5941	6500	0.9696
0.4341	9.9631	6750	0.9671

Framework versions

• PEFT 0.10.0
• Transformers 4.40.0
• Pytorch 2.2.2+cu121
• Datasets 2.19.0
• Tokenizers 0.19.1