Hanhpt23/whisper-medium-chinesemed-full

openai/whisper-medium

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

• Loss: 1.1592
• Wer: 80.9430

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
0.6574	1.0	161	0.8255	104.1257
0.3457	2.0	322	0.8750	90.9627
0.2273	3.0	483	0.9033	84.4794
0.1392	4.0	644	0.9562	81.3360
1.8629	5.0	805	1.3537	127.1120
0.0816	6.0	966	1.0682	88.0157
0.0722	7.0	1127	1.0343	100.0
0.0494	8.0	1288	1.0605	84.4794
0.0405	9.0	1449	1.0960	86.6405
0.0403	10.0	1610	1.1383	87.4263
0.0216	11.0	1771	1.0693	88.9980
0.0166	12.0	1932	1.0588	88.0157
0.0126	13.0	2093	1.1074	79.1749
0.0083	14.0	2254	1.1398	83.6935
0.0075	15.0	2415	1.1451	88.6051
0.0019	16.0	2576	1.1521	85.0688
0.0006	17.0	2737	1.1442	81.1395
0.0007	18.0	2898	1.1521	82.3183
0.0003	19.0	3059	1.1581	80.9430
0.0002	20.0	3220	1.1592	80.9430

Framework versions

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