whisper-fine-tune

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: 1.2916
• Wer: 0.8706

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: 1e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 400
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.776	1.0	19	1.6462	0.9882
1.5438	2.0	38	1.5323	0.9265
1.2294	3.0	57	1.2822	0.9118
0.9329	4.0	76	1.0808	0.9029
0.6652	5.0	95	1.1272	0.9029
0.4095	6.0	114	0.9926	0.8824
0.2196	7.0	133	1.0875	0.8912
0.109	8.0	152	1.2208	0.8735
0.053	9.0	171	1.1724	0.8735
0.0455	10.0	190	1.2695	0.8941
0.0381	11.0	209	1.2097	0.8794
0.0357	12.0	228	1.2848	0.8824
0.024	13.0	247	1.3479	0.8824
0.0301	14.0	266	1.1870	0.8794
0.0218	15.0	285	1.3447	0.8941
0.0274	16.0	304	1.3277	0.8971
0.0295	17.0	323	1.3363	0.8794
0.0266	18.0	342	1.2948	0.8882
0.0296	19.0	361	1.1923	0.8912
0.0246	20.0	380	1.2975	0.8941
0.0172	21.0	399	1.3196	0.9
0.0287	22.0	418	1.2178	0.8882
0.0277	23.0	437	1.2133	0.8882
0.022	24.0	456	1.2898	0.8794
0.0112	25.0	475	1.4235	0.8735
0.0086	26.0	494	1.3367	0.8735
0.0072	27.0	513	1.3071	0.8706
0.0042	28.0	532	1.3378	0.8647
0.0021	29.0	551	1.3468	0.8706
0.0014	30.0	570	1.2916	0.8706

Framework versions

• Transformers 4.57.3
• Pytorch 2.9.0+cu126
• Datasets 4.0.0
• Tokenizers 0.22.1