Curiousfox/whisper_new_ver3

Hokkien-to-Tai Lo Whisper ver 3

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 16.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5026
• Wer: 82.8578

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-06
• train_batch_size: 8
• eval_batch_size: 16
• seed: 42
• 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: 1000
• training_steps: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.9959	0.2581	800	1.1302	99.9772
0.8265	0.5161	1600	0.7367	92.8555
0.6895	0.7742	2400	0.6409	87.4001
0.5814	1.0323	3200	0.5745	85.9849
0.4864	1.2903	4000	0.5512	85.4143
0.4662	1.5484	4800	0.5334	85.8480
0.4421	1.8065	5600	0.5154	83.6339
0.449	2.0645	6400	0.5097	83.3600
0.3807	2.3226	7200	0.5058	83.0861
0.3783	2.5806	8000	0.5026	82.8578

Framework versions

• Transformers 4.49.0
• Pytorch 2.5.1+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0