jeff7522553/whisper-medium-tvgh

Whisper medium TW - for tvgh

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

• Loss: 0.0540
• Cer: 12.2258

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: 1
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 2
• 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: 6000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.173	0.3259	1000	0.1400	13.8168
0.0526	0.6519	2000	0.0940	9.6739
0.1043	0.9778	3000	0.0752	15.4866
0.0335	1.3038	4000	0.0716	8.8233
0.0717	1.6297	5000	0.0597	9.9968
0.0422	1.9557	6000	0.0540	12.2258

Framework versions

• Transformers 4.50.3
• Pytorch 2.9.1+cu128
• Datasets 4.4.1
• Tokenizers 0.21.4