tylee4/whisper-tiny-ko

Whisper Tiny Ko - 0.1

This model is a fine-tuned version of openai/whisper-tiny on the zeroth_korean dataset. It achieves the following results on the evaluation set:

• Loss: 0.1870
• Cer: 9.5399

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: 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: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.4034	0.7184	1000	0.4414	14.7103
0.2432	1.4368	2000	0.3385	11.8208
0.1511	2.1552	3000	0.2778	9.6734
0.1295	2.8736	4000	0.2440	8.1855
0.0905	3.5920	5000	0.2219	7.8321
0.0638	4.3103	6000	0.2071	8.2915
0.0443	5.0287	7000	0.1970	9.0766
0.0534	5.7471	8000	0.1914	8.1069
0.0503	6.4655	9000	0.1879	6.7682
0.0308	7.1839	10000	0.1870	9.5399

Framework versions

• Transformers 4.53.0
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.2