CheeLi03/whisper-tiny-ja-puct-combine-8k

Whisper Tiny Japanese Combine 4k - Chee Li

This model is a fine-tuned version of openai/whisper-tiny on the Meta JSON Japanese Dataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.8869
• Wer: 396.6874

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use 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: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.441	4.1322	1000	2.4726	406.5217
1.8098	8.2645	2000	2.0185	462.4224
1.2666	12.3967	3000	1.5918	404.3478
0.8324	16.5289	4000	1.2738	460.8696
0.5744	20.6612	5000	1.0687	607.0393
0.3308	24.7934	6000	0.9561	532.7122
0.242	28.9256	7000	0.9024	461.0766
0.1651	33.0579	8000	0.8869	396.6874

Framework versions

• Transformers 4.46.2
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.20.1