refactor(handler): charge paresseusement le modèle fine-tuné pour éviter les erreurs à l’import

app/handler.py

@@ -1,17 +1,19 @@
 from models.zero_shot import ZeroShotModel
 from models.few_shot import FewShotModel
-from models.fine_tuned import FineTunedModel
 
 zero_shot_model = ZeroShotModel()
 few_shot_model = FewShotModel()
-fine_tuned_model = FineTunedModel()
+
+def get_fine_tuned_model():
+    from models.fine_tuned import FineTunedModel
+    return FineTunedModel()
 
 def predict(text: str, model_type: str = "zero-shot") -> str:
     if model_type == "few-shot":
         results = few_shot_model.predict(text)
         title = "Few-Shot"
     elif model_type == "fine-tuned":
-        results = fine_tuned_model.predict(text)
+        results = get_fine_tuned_model().predict(text)
         title = "Fine-Tuned"
     else:
         results = zero_shot_model.predict(text)
@@ -19,5 +21,5 @@ def predict(text: str, model_type: str = "zero-shot") -> str:
 
     output = f"### Résultat de la classification ({title}) :\n\n"
     for label, score in results:
-        output += f"- **{label}** : {score*100:.1f}%\n"
+        output += f"- **{label}** : {score * 100:.1f}%\n"
     return output