whisper-medium-Split-Sentences-cleanpunc

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

• Loss: 0.6916
• Cer: 15.8237
• Wer: 27.8527

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: 5e-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
2.5398	1.0	1353	0.6961	33.6482	54.1036
0.9355	2.0	2706	0.6405	34.2321	51.6942
0.7293	3.0	4059	0.5997	28.1857	44.6026
0.6003	4.0	5412	0.5919	26.6651	42.5012
0.5051	5.0	6765	0.5781	26.4160	40.6325
0.4287	6.0	8118	0.5792	22.4282	36.1832
0.3646	7.0	9471	0.5839	19.7198	32.6853
0.3112	8.0	10824	0.5972	18.8569	31.7544
0.2648	9.0	12177	0.6061	19.4667	32.7127
0.2258	10.0	13530	0.6073	19.1717	33.1166
0.1928	11.0	14883	0.6155	16.8899	29.5092
0.1649	12.0	16236	0.6273	17.8584	30.6044
0.1409	13.0	17589	0.6338	17.9501	30.4401
0.1219	14.0	18942	0.6458	17.9700	30.4128
0.1047	15.0	20295	0.6494	17.1589	29.6530
0.0906	16.0	21648	0.6551	17.1569	29.4065
0.0803	17.0	23001	0.6580	15.9931	27.9211
0.0701	18.0	24354	0.6706	16.2163	28.3045
0.0621	19.0	25707	0.6736	16.3358	28.3798
0.0561	20.0	27060	0.6802	16.5411	28.6125
0.0508	21.0	28413	0.6810	16.0489	28.0649
0.0463	22.0	29766	0.6884	16.1525	28.2223
0.0434	23.0	31119	0.6916	15.8237	27.8527
0.0409	24.0	32472	0.6930	16.0768	28.0101
0.0393	25.0	33825	0.6935	16.3697	28.2429

Framework versions

• Transformers 4.53.3
• Pytorch 2.7.1+cu118
• Datasets 3.6.0
• Tokenizers 0.21.2