nkkbr/whisper-large-v3-zatoichi-ja-EX-4

Whisper Large v3 - Japanese Zatoichi ASR

This model is a fine-tuned version of openai/whisper-large-v3 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5821
• Wer: 63.7025

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: 32
• eval_batch_size: 8
• seed: 42
• optimizer: Use 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_ratio: 0.1
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.018	0.1471	5	1.0820	93.8561
0.8059	0.2941	10	0.8298	128.1326
0.7047	0.4412	15	0.7087	77.2029
0.5925	0.5882	20	0.6603	83.5085
0.548	0.7353	25	0.6197	76.7987
0.7339	0.8824	30	0.5944	80.6791
0.3835	1.0294	35	0.5788	72.9184
0.3677	1.1765	40	0.5788	80.9216
0.379	1.3235	45	0.5612	74.2118
0.3885	1.4706	50	0.5489	74.7777
0.3784	1.6176	55	0.5379	74.9394
0.414	1.7647	60	0.5325	64.5918
0.3427	1.9118	65	0.5276	63.6217
0.2165	2.0588	70	0.5279	63.7833
0.1962	2.2059	75	0.5491	61.0348
0.2305	2.3529	80	0.5551	64.0259
0.1687	2.5	85	0.5551	61.6815
0.2256	2.6471	90	0.5582	68.0679
0.2049	2.7941	95	0.5580	68.8763
0.2029	2.9412	100	0.5597	69.6039
0.1489	3.0882	105	0.5674	64.2684
0.1794	3.2353	110	0.5713	67.5020
0.1124	3.3824	115	0.5821	71.0590
0.1464	3.5294	120	0.5850	63.2983
0.1253	3.6765	125	0.5851	66.6936
0.1323	3.8235	130	0.5835	66.4511
0.1083	3.9706	135	0.5821	63.7025

Framework versions

• Transformers 4.57.3
• Pytorch 2.9.1+cu128
• Datasets 4.4.1
• Tokenizers 0.22.1