Hanhpt23/whisper-small-engmed-free_ED0-8

openai/whisper-small

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

• Loss: 0.0002
• Wer: 16.9843
• Cer: 13.2128

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	Cer
0.7745	1.0	386	0.3281	23.8481	18.9940
0.4222	2.0	772	0.1559	34.1276	31.2181
0.2229	3.0	1158	0.0890	41.8389	41.5395
0.0959	4.0	1544	0.0530	43.5633	41.1019
0.0783	5.0	1930	0.0374	35.9313	35.5561
0.0491	6.0	2316	0.0241	31.5489	24.3995
0.0384	7.0	2702	0.0182	26.3650	22.6028
0.0454	8.0	3088	0.0122	25.1452	20.9192
0.0153	9.0	3474	0.0076	21.1914	17.8663
0.013	10.0	3860	0.0059	20.7264	16.9106
0.0049	11.0	4246	0.0035	18.8548	14.8285
0.0028	12.0	4632	0.0026	16.8447	12.9932
0.0045	13.0	5018	0.0017	20.3620	16.0350
0.0015	14.0	5404	0.0010	18.4088	13.8127
0.0021	15.0	5790	0.0006	16.3526	13.1850
0.0042	16.0	6176	0.0004	16.0627	12.6793
0.0009	17.0	6562	0.0003	16.4933	13.0223
0.0003	18.0	6948	0.0002	16.9122	12.9176
0.0006	19.0	7334	0.0002	16.8802	13.1395
0.0001	20.0	7720	0.0002	16.9843	13.2128

Framework versions

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