jethrowang/vanilla-whisper-medium_hat

Whisper Medium Hakka Condenser

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

• Loss: 0.1958
• Cer: 7.8843

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 16
• 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: 3904
• training_steps: 39045
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.0115	15.9711	7809	0.2071	9.7926
0.0044	31.9403	15618	0.1995	9.3661
0.0022	47.9096	23427	0.2186	9.5395
0.0003	63.8789	31236	0.1999	8.8980
0.0	79.8481	39045	0.1958	7.8843

Framework versions

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