bagasshw/whisper-large-v2-jv-filtered

This model is a fine-tuned version of openai/whisper-large-v2 on the jv_id_asr_split dataset. It achieves the following results on the evaluation set:

• Loss: 0.0756
• Wer: 6.3496

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Use OptimizerNames.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: 60000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2396	0.1535	5000	0.2416	19.2376
0.1912	0.3071	10000	0.2009	16.2781
0.1744	0.4606	15000	0.1687	14.4586
0.1448	0.6142	20000	0.1494	12.9593
0.138	0.7677	25000	0.1333	11.8420
0.1312	0.9213	30000	0.1192	10.7587
0.0605	1.0748	35000	0.1096	9.9808
0.0524	1.2284	40000	0.1022	9.3473
0.0506	1.3819	45000	0.0929	8.2091
0.0476	1.5355	50000	0.0859	7.7627
0.0389	1.6890	55000	0.0795	6.7298
0.0357	1.8426	60000	0.0756	6.3496

Framework versions

• Transformers 4.50.0.dev0
• Pytorch 2.7.0+cu128
• Datasets 2.18.0
• Tokenizers 0.21.1