README.md

metadata

library_name: transformers
language:
  - ko
license: mit
base_model: openai/whisper-large-v3-turbo
tags:
  - generated_from_trainer
datasets:
  - Bingsu/zeroth-korean
  - mozilla-foundation/common_voice_17_0
metrics:
  - wer
model-index:
  - name: Whisper large v3 turbo Korean - imTak
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Zeroth-Korean
          type: Bingsu/zeroth-korean
          args: 'config: ko, split: test'
        metrics:
          - name: Wer
            type: wer
            value: 5.270290618882698

Whisper large v3 turbo Korean - imTak

This model is a fine-tuned version of imTak/whisper_large_v3_ko_ft on the Zeroth-Korean dataset. It achieves the following results on the evaluation set:

• Loss: 0.0670
• Wer: 5.2703

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: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1068	0.7184	1000	0.1216	8.6132
0.0388	1.4368	2000	0.0905	5.3606
0.0089	2.1552	3000	0.0707	4.7282
0.0082	2.8736	4000	0.0670	5.2703

Framework versions

• Transformers 4.45.0
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3