metadata
language:
- ko
license: apache-2.0
library_name: peft
tags:
- generated_from_trainer
base_model: openai/whisper-large-v2
datasets:
- customd_ataset
model-index:
- name: Whisper large-v2 Korean - ML_project_custom_data_5epoch_with500
results: []
Whisper large-v2 Korean - ML_project_custom_data_5epoch_with500
This model is a fine-tuned version of openai/whisper-large-v2 on the customd_ataset dataset. It achieves the following results on the evaluation set:
- Loss: 0.7892
- Cer: 80.2045
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: 0.001
- train_batch_size: 4
- eval_batch_size: 2
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 50
- num_epochs: 5
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer |
|---|---|---|---|---|
| 0.114 | 1.0 | 113 | 0.8207 | 86.1943 |
| 0.1328 | 2.0 | 226 | 0.8307 | 96.2016 |
| 0.0678 | 3.0 | 339 | 0.8054 | 96.7860 |
| 0.0246 | 4.0 | 452 | 0.7812 | 78.8897 |
| 0.0155 | 5.0 | 565 | 0.7892 | 80.2045 |
Framework versions
- PEFT 0.11.2.dev0
- Transformers 4.41.2
- Pytorch 2.3.0+cu121
- Datasets 2.19.2
- Tokenizers 0.19.1