whisper-small-kh-v3

This model is a fine-tuned version of openai/whisper-tiny on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4105
• Wer: 76.8923

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 80
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.2514	1.2300	500	1.2510	118.6532
1.0397	2.4600	1000	1.0474	119.5580
0.7959	3.6900	1500	0.8172	110.5968
0.6345	4.9200	2000	0.6525	102.7667
0.5037	6.1501	2500	0.5592	90.0992
0.4343	7.3801	3000	0.5061	89.5076
0.3917	8.6101	3500	0.4723	84.3223
0.3718	9.8401	4000	0.4521	82.5126
0.3274	11.0701	4500	0.4379	79.6242
0.3	12.3001	5000	0.4289	80.8074
0.2786	13.5301	5500	0.4216	79.2935
0.2707	14.7601	6000	0.4156	77.9711
0.2732	15.9902	6500	0.4138	76.4399
0.2591	17.2202	7000	0.4123	76.6835
0.2538	18.4502	7500	0.4114	77.0663
0.2458	19.6802	8000	0.4105	76.8923

Framework versions

• Transformers 4.44.2
• Pytorch 2.8.0+cu126
• Datasets 4.2.0
• Tokenizers 0.19.1