JunWorks/whisper-small-zhTW

Whisper Small zh-TW

This model is a fine-tuned version of openai/whisper-small on the Common Voice 13.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2504

• Cer: 10.4045

• But when loaded CER = 10.3295

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: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.0898	1.38	1000	0.1981	11.5149
0.0283	2.75	2000	0.1991	10.3994
0.0037	4.13	3000	0.2141	10.5107
0.0038	5.5	4000	0.2237	10.4252
0.0033	6.88	5000	0.2263	10.3062
0.0004	8.25	6000	0.2339	10.8238
0.0003	9.63	7000	0.2400	10.4252
0.0002	11.0	8000	0.2451	10.5262
0.0002	12.38	9000	0.2487	10.3605
0.0002	13.76	10000	0.2504	10.4045

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2