File size: 1,934 Bytes
e11c249 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 |
# Marian Fine-tuned English-French Translation Model
## Model Description
This model is a fine-tuned version of `Helsinki-NLP/opus-mt-en-fr`, specifically trained for English to French translation. The base model was further trained on the `KDE4` dataset to improve translation quality for technical and software-related content.
## Model Training Details
### Training Dataset
- **Dataset**: KDE4 Dataset (English-French parallel corpus)
- **Split Distribution**:
- Training set: 189,155 examples (90%)
- Test set: 21,018 examples (10%)
### Training Configuration
- **Base Model**: Helsinki-NLP/opus-mt-en-fr
- **Training Arguments**:
- Learning rate: 2e-5
- Batch size: 32 (training), 64 (evaluation)
- Number of epochs: 10
- Weight decay: 0.01
- FP16 training enabled
- Evaluation strategy: Before and after training
- Checkpoint saving: Every epoch (maximum 3 saved)
- Training device: GPU with mixed precision (fp16)
## Model Results
### Evaluation Metrics
The model was evaluated using the BLEU score. The evaluation results before and after training are summarized in the table below:
| **Stage** | **Eval Loss** | **BLEU Score** |
|--------------------|---------------|----------------|
| **Before Training** | 1.700 | 38.97 |
| **After Training** | 0.796 | 54.96 |
### Training Loss
The training loss decreased over the epochs, indicating that the model was learning effectively. The final training loss was approximately 0.710.
## Model Usage
```python
from transformers import pipeline
model_checkpoint = "Prikshit7766/marian-finetuned-kde4-en-to-fr"
translator = pipeline("translation", model=model_checkpoint)
translator("Default to expanded threads")
```
### Example Output
```plaintext
[{'translation_text': 'Par défaut, développer les fils de discussion'}]
```
|