|
|
--- |
|
|
language: fr |
|
|
license: mit |
|
|
tags: |
|
|
- nlp |
|
|
- text-classification |
|
|
- sentiment-analysis |
|
|
- french |
|
|
base_model: asi/gpt-fr-cased-small |
|
|
datasets: |
|
|
- allocine |
|
|
model_name: gpt-fr-sentiment-classifier |
|
|
widget: |
|
|
- text: "Ce film est une véritable réussite !" |
|
|
metrics: |
|
|
- accuracy |
|
|
--- |
|
|
|
|
|
# GPT-FR – Classifieur de sentiment (Allociné) |
|
|
|
|
|
Modèle binaire **positif / négatif** affiné sur le jeu de données |
|
|
**Allociné** (160 000 critiques de films en français). |
|
|
La tête de classification repose sur le **GPT-2 francophone** |
|
|
`asi/gpt-fr-cased-small`. |
|
|
|
|
|
| | Valeur | |
|
|
|------------------------|--------| |
|
|
| **Base** | asi/gpt-fr-cased-small | |
|
|
| **Dataset** | Allociné (train 160 k / val 20 k) | |
|
|
| **Max seq length** | 128 | |
|
|
| **Batch size** | 8 | |
|
|
| **Époques** | 2 | |
|
|
| **Accuracy (val)** | **94 %** | |
|
|
|
|
|
## Exemple d’utilisation |
|
|
|
|
|
```python |
|
|
from transformers import pipeline |
|
|
|
|
|
sentiment = pipeline( |
|
|
"text-classification", |
|
|
model="mkd13mkd/gpt-fr-sentiment-classifier", |
|
|
tokenizer="asi/gpt-fr-cased-small" |
|
|
) |
|
|
|
|
|
print(sentiment("Ce film est une véritable réussite !")) |
|
|
# [{'label': 'LABEL_1', 'score': 0.97}] → LABEL_1 = positif |
|
|
|
