whisper-medium-ml-exp2

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: 0.2220
• Wer: 57.6922

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: 16
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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: 500
• training_steps: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3119	0.0333	500	0.4193	79.0917
0.0405	0.0667	1000	0.4335	73.5830
0.0355	0.1	1500	0.4332	77.2785
0.126	0.1333	2000	0.1967	58.4021
0.0519	0.1667	2500	0.1861	58.5671
0.0439	0.2	3000	0.1942	57.4274
0.0534	1.0112	3500	0.1936	61.1497
0.0214	1.0445	4000	0.2253	59.7816
0.0129	1.0779	4500	0.2630	61.0614
0.048	1.1112	5000	0.1780	56.3606
0.047	1.1445	5500	0.1638	52.9951
0.0325	1.1779	6000	0.1683	54.5512
0.0293	1.2112	6500	0.1689	57.2451
0.028	2.0224	7000	0.2145	56.5237
0.009	2.0557	7500	0.2227	56.3068
0.0076	2.0891	8000	0.2750	66.0540
0.0385	2.1224	8500	0.2178	54.4514
0.0245	2.1557	9000	0.1721	52.0031
0.0226	2.1891	9500	0.1741	53.7511
0.0212	3.0003	10000	0.2001	56.1495
0.0121	3.0336	10500	0.2322	55.4722
0.0042	3.0669	11000	0.2403	57.6864
0.0059	3.1003	11500	0.2953	64.0067
0.0248	3.1336	12000	0.1744	51.3412
0.0172	3.1669	12500	0.1872	53.5324
0.015	3.2003	13000	0.1930	54.7028
0.0158	4.0115	13500	0.2173	60.9636
0.0028	4.0448	14000	0.2330	53.4921
0.0028	4.0781	14500	0.2415	53.4767
0.0194	4.1115	15000	0.2220	57.6922

Framework versions

• Transformers 4.51.1
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1