Create README.md
Browse files
README.md
CHANGED
|
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# DistilBERT Query Classifier
|
| 2 |
+
|
| 3 |
+
Modèle de classification binaire pour distinguer les requêtes RAG des demandes d'envoi de messages.
|
| 4 |
+
|
| 5 |
+
## Utilisation
|
| 6 |
+
|
| 7 |
+
```python
|
| 8 |
+
from transformers import pipeline
|
| 9 |
+
|
| 10 |
+
# Charger le modèle
|
| 11 |
+
classifier = pipeline("text-classification", model="your-username/distilbert-query-classifier")
|
| 12 |
+
|
| 13 |
+
# Classifier une requête
|
| 14 |
+
result = classifier("What are the prerequisites for the machine learning course?")
|
| 15 |
+
print(result)
|
| 16 |
+
# [{'label': 'question_rag', 'score': 0.92}]
|
| 17 |
+
```
|
| 18 |
+
|
| 19 |
+
## Classes
|
| 20 |
+
|
| 21 |
+
- **question_rag** (0): Questions nécessitant une recherche RAG
|
| 22 |
+
- **send_message** (1): Demandes d'envoi de messages
|
| 23 |
+
|
| 24 |
+
## Exemples
|
| 25 |
+
|
| 26 |
+
```python
|
| 27 |
+
queries = [
|
| 28 |
+
"What topics are covered in the Python course?", # → question_rag
|
| 29 |
+
"Send a message to John about the meeting", # → send_message
|
| 30 |
+
]
|
| 31 |
+
|
| 32 |
+
results = classifier(queries)
|
| 33 |
+
```
|
| 34 |
+
|
| 35 |
+
## Détails techniques
|
| 36 |
+
|
| 37 |
+
- **Modèle**: distilbert-base-uncased
|
| 38 |
+
- **Dataset**: 98 exemples (50/50 split)
|
| 39 |
+
- **Accuracy**: 93% sur test set
|
| 40 |
+
- **Couches entraînées**: 2 dernières couches + classifier
|
| 41 |
+
- **Epochs**: 10
|
| 42 |
+
|
| 43 |
+
## Limitations
|
| 44 |
+
|
| 45 |
+
- Petit dataset d'entraînement (98 exemples)
|
| 46 |
+
- Anglais uniquement
|
| 47 |
+
- Classification binaire seulement
|