jethrowang/vanilla-whisper-small_hat

Whisper Small Hakka Condenser

This model is a fine-tuned version of openai/whisper-small on the HAT ASR Aligned dataset. It achieves the following results on the evaluation set:

• Loss: 0.1313
• Cer: 6.1701

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• 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: 1366
• training_steps: 13664
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.0405	3.9933	1952	0.1910	14.4081
0.0121	7.9852	3904	0.1850	9.9556
0.0061	11.9770	5856	0.1636	8.6610
0.0032	15.9688	7808	0.1563	7.6207
0.0004	19.9606	9760	0.1421	6.5978
0.0001	23.9524	11712	0.1332	6.4117
0.0	27.9442	13664	0.1313	6.1701

Framework versions

• Transformers 4.49.0
• Pytorch 2.0.0.post304
• Datasets 3.3.2
• Tokenizers 0.21.0