nkkbr/whisper-large-v3-zatoichi-ja-20251210_031911

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: 0.5611
• Wer: 71.4678

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: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7135	0.1678	25	0.8879	89.9177
0.6885	0.3356	50	0.7025	83.8820
0.6108	0.5034	75	0.6615	77.7778
0.4705	0.6711	100	0.6191	76.6118
0.6904	0.8389	125	0.6027	80.9328
0.4674	1.0067	150	0.5799	72.6337
0.4163	1.1745	175	0.5956	75.1715
0.2719	1.3423	200	0.5803	75.7202
0.2796	1.5101	225	0.5703	72.4966
0.4221	1.6779	250	0.5676	72.3594
0.3267	1.8456	275	0.5611	71.4678

Framework versions

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