Jobaula/whisper-medium-nan-tw-common-voice

Whisper medium nan-tw common voice

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

• Loss: 0.0141
• Model Preparation Time: 0.0121
• Wer: 0.9615
• Cer: 0.9524

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: 2
• eval_batch_size: 2
• seed: 42
• optimizer: Use adamw_bnb_8bit 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: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer	Cer
0.97	0.2	1000	0.7356	0.0121	38.1731	38.4762
0.3044	1.0388	2000	0.3099	0.0121	23.4615	23.9048
0.3108	1.2388	3000	0.1153	0.0121	7.5	7.7143
0.0544	2.0776	4000	0.0295	0.0121	2.3077	2.2857
0.0678	2.2776	5000	0.0141	0.0121	0.9615	0.9524

Framework versions

• Transformers 4.47.0.dev0
• Pytorch 2.5.1+cu121
• Datasets 3.1.0
• Tokenizers 0.20.3