exp_003_base_frozen

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

• Loss: 0.1331
• Wer: 35.9827
• Wer Ortho: 38.8106
• Cer: 13.3048

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: 2e-05
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• 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: 1000
• training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Wer Ortho	Cer
3.3664	0.6410	500	0.7780	70.2985	73.1122	21.4014
2.4777	1.2821	1000	0.5469	56.7788	60.0050	17.5319
2.0633	1.9231	1500	0.4300	48.2554	51.8306	15.0486
1.6801	2.5641	2000	0.3588	44.1239	47.6042	14.2280
1.3374	3.2051	2500	0.3044	40.7650	44.2131	13.0829
1.2080	3.8462	3000	0.2580	39.0267	42.3347	12.9674
0.9522	4.4872	3500	0.2196	37.2339	40.5644	12.4938
0.7244	5.1282	4000	0.1886	37.2423	40.4023	13.0369
0.6902	5.7692	4500	0.1677	36.5789	39.5254	12.6352
0.5293	6.4103	5000	0.1495	36.4404	39.1639	12.6400
0.4483	7.0513	5500	0.1331	35.9827	38.8106	13.3048
0.3833	7.6923	6000	0.1220	36.6251	39.3841	13.2253
0.2814	8.3333	6500	0.1151	36.9526	39.7124	13.4726
0.2677	8.9744	7000	0.1067	36.5159	39.1930	14.1547
0.1995	9.6154	7500	0.1029	37.5698	40.3358	14.3636
0.1543	10.2564	8000	0.1013	37.2969	39.9493	13.7012

Framework versions

• Transformers 5.0.0
• Pytorch 2.10.0+cu128
• Datasets 3.6.0
• Tokenizers 0.22.2