exp_001_base_vanilla

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: 0.0281
• Wer: 21.2201
• Wer Ortho: 22.5616
• Cer: 8.5150

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: 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
2.6093	0.6410	500	0.5502	55.5654	59.1115	17.0550
1.8138	1.2821	1000	0.3793	44.3549	47.6208	14.6009
1.3138	1.9231	1500	0.2431	41.7559	44.2339	16.5479
0.7587	2.5641	2000	0.1512	33.0562	35.4943	11.5486
0.3661	3.2051	2500	0.1045	29.6133	31.8373	10.6197
0.3351	3.8462	3000	0.0754	30.1045	32.1406	11.4877
0.1857	4.4872	3500	0.0664	28.7568	30.6155	10.9869
0.0939	5.1282	4000	0.0549	28.3327	30.1542	10.8709
0.1039	5.7692	4500	0.0493	28.3159	30.0586	10.8963
0.0600	6.4103	5000	0.0437	26.5819	28.2966	10.3570
0.0486	7.0513	5500	0.0432	26.7708	28.4628	10.5080
0.0351	7.6923	6000	0.0388	25.8219	27.5111	10.2674
0.0212	8.3333	6500	0.0369	24.9276	26.6301	9.8892
0.0189	8.9744	7000	0.0340	25.5574	27.0789	10.0536
0.0117	9.6154	7500	0.0342	23.9829	25.5995	9.6716
0.0054	10.2564	8000	0.0310	23.1473	24.5439	9.1223
0.0038	10.8974	8500	0.0290	22.4126	23.8457	9.0356
0.0011	11.5385	9000	0.0285	21.6904	23.0063	8.6243
0.0005	12.1795	9500	0.0283	21.4847	22.8525	8.5869
0.0004	12.8205	10000	0.0281	21.2201	22.5616	8.5150

Framework versions

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