metadata
library_name: transformers
language:
- ko
license: mit
base_model: imTak/whisper_large_v3_turbo_Korean2
tags:
- generated_from_trainer
datasets:
- imTak/Economy
metrics:
- wer
model-index:
- name: Whisper large v3 turbo Korean-Economy
results:
- task:
name: Automatic Speech Recognition
type: automatic-speech-recognition
dataset:
name: Economy
type: imTak/Economy
args: 'config: ko, split: test'
metrics:
- name: Wer
type: wer
value: 44.99209128911987
Whisper large v3 turbo Korean-Economy
This model is a fine-tuned version of imTak/whisper_large_v3_ko_ft_ft on the Economy dataset. It achieves the following results on the evaluation set:
- Loss: 0.7148
- Wer: 44.9921
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: 32
- eval_batch_size: 32
- 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: 8000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.8263 | 0.4630 | 1000 | 0.8210 | 49.0241 |
| 0.7497 | 0.9259 | 2000 | 0.7351 | 47.6006 |
| 0.4979 | 1.3889 | 3000 | 0.6992 | 45.6375 |
| 0.5197 | 1.8519 | 4000 | 0.6659 | 44.3410 |
| 0.4264 | 2.3148 | 5000 | 0.7168 | 46.6459 |
| 0.3911 | 2.7778 | 6000 | 0.6988 | 45.0726 |
| 0.2565 | 3.2407 | 7000 | 0.7203 | 44.8000 |
| 0.2462 | 3.7037 | 8000 | 0.7148 | 44.9921 |
Framework versions
- Transformers 4.45.0
- Pytorch 2.5.1+cu124
- Datasets 3.1.0
- Tokenizers 0.20.3