Hanhpt23/whisper-small-chinesemed-full

openai/whisper-small

This model is a fine-tuned version of openai/whisper-small on the pphuc25/ChiMed dataset. It achieves the following results on the evaluation set:

• Loss: 1.1524
• Wer: 82.7112

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: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6099	1.0	161	0.7541	214.5383
0.3252	2.0	322	0.8393	166.9941
0.1959	3.0	483	0.8690	116.3065
0.101	4.0	644	0.9098	87.8193
0.0746	5.0	805	0.9271	85.8546
0.0384	6.0	966	0.9985	84.0864
0.0462	7.0	1127	1.0526	80.3536
0.0415	8.0	1288	1.0810	84.2829
0.0326	9.0	1449	1.1428	82.9077
0.023	10.0	1610	1.1043	84.0864
0.0157	11.0	1771	1.0950	81.7289
0.0152	12.0	1932	1.1084	83.4971
0.0068	13.0	2093	1.0879	81.1395
0.0056	14.0	2254	1.1089	83.6935
0.0012	15.0	2415	1.1175	82.3183
0.0005	16.0	2576	1.1441	84.2829
0.0005	17.0	2737	1.1475	83.1041
0.0008	18.0	2898	1.1505	82.7112
0.0002	19.0	3059	1.1518	82.7112
0.0002	20.0	3220	1.1524	82.7112

Framework versions

• Transformers 4.41.1
• Pytorch 2.3.0
• Datasets 2.19.1
• Tokenizers 0.19.1