nkkbr/whisper-large-v3-zatoichi-ja-EX-3

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: 0.5654
• Wer: 67.4590

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
1.1025	0.1220	5	1.0834	114.6721
0.9692	0.2439	10	0.8239	132.3770
0.7425	0.3659	15	0.6744	100.2459
0.6636	0.4878	20	0.6218	71.6393
0.638	0.6098	25	0.5902	82.2951
0.5479	0.7317	30	0.5644	67.4590
0.5902	0.8537	35	0.5436	68.3607
0.5849	0.9756	40	0.5308	72.2951
0.4034	1.0976	45	0.5174	77.8689
0.4465	1.2195	50	0.5231	62.7049
0.3736	1.3415	55	0.5224	61.9672
0.4433	1.4634	60	0.5222	62.4590
0.4117	1.5854	65	0.5141	65.0820
0.3197	1.7073	70	0.5030	72.7049
0.426	1.8293	75	0.5044	64.9180
0.3499	1.9512	80	0.5026	72.8689
0.2099	2.0732	85	0.5007	70.9016
0.2632	2.1951	90	0.5099	63.0328
0.1636	2.3171	95	0.5216	61.5574
0.2176	2.4390	100	0.5345	59.8361
0.2172	2.5610	105	0.5289	69.0984
0.2307	2.6829	110	0.5262	69.5902
0.2131	2.8049	115	0.5262	69.9180
0.2044	2.9268	120	0.5266	66.8033
0.1408	3.0488	125	0.5287	60.7377
0.1479	3.1707	130	0.5427	67.9508
0.1395	3.2927	135	0.5611	66.3115
0.1244	3.4146	140	0.5647	61.6393
0.1205	3.5366	145	0.5642	61.2295
0.145	3.6585	150	0.5665	62.1311
0.1297	3.7805	155	0.5653	61.9672
0.1534	3.9024	160	0.5654	67.4590

Framework versions

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