Spaces:
Sleeping
Sleeping
| from typing import List, Dict, Literal | |
| import json | |
| from langchain_community.vectorstores import FAISS | |
| from langchain_openai import OpenAIEmbeddings | |
| from langchain_openai import ChatOpenAI | |
| from langchain.prompts import ChatPromptTemplate | |
| from app.core.config import settings | |
| from app.models.quiz import Question | |
| import logging | |
| logger = logging.getLogger(__name__) | |
| class QuizGenerator: | |
| def __init__(self, provider: Literal["openai", "deepseek"] = "openai"): | |
| self.embeddings = OpenAIEmbeddings(openai_api_key=settings.OPENAI_API_KEY) | |
| self.vector_store = FAISS.load_local( | |
| settings.VECTOR_DB_PATH, | |
| self.embeddings, | |
| allow_dangerous_deserialization=True | |
| ) | |
| # Configuration selon le provider | |
| if provider == "deepseek": | |
| self.llm = ChatOpenAI( | |
| model_name="deepseek-chat", | |
| temperature=0.7, | |
| openai_api_key=settings.DEEPSEEK_API_KEY, | |
| base_url="https://api.deepseek.com" | |
| ) | |
| else: # openai | |
| self.llm = ChatOpenAI( | |
| model_name=settings.MODEL_NAME, | |
| temperature=0.7, | |
| openai_api_key=settings.OPENAI_API_KEY | |
| ) | |
| def clean_json_response(self, response_text: str) -> str: | |
| """Nettoie la réponse du LLM pour obtenir un JSON valide.""" | |
| # Supprime les backticks markdown et le mot 'json' | |
| cleaned = response_text.replace('```json', '').replace('```', '').strip() | |
| # Supprime les espaces et sauts de ligne superflus au début et à la fin | |
| return cleaned.strip() | |
| async def generate_quiz(self, theme: str, num_questions: int = 5) -> List[Question]: | |
| logger.debug(f"Génération de quiz - Thème: {theme}, Nb questions: {num_questions}") | |
| try: | |
| # Récupérer le contexte pertinent | |
| docs = self.vector_store.similarity_search(theme, k=3) | |
| context = "\n".join([doc.page_content for doc in docs]) | |
| # Template pour la génération de questions | |
| prompt = ChatPromptTemplate.from_template(""" | |
| Tu es un générateur de quiz intelligent. | |
| Si tu as du contexte pertinent, utilise-le : {context} | |
| Sinon, utilise tes connaissances générales. | |
| Génère {num_questions} questions de quiz sur le thème: {theme} | |
| IMPORTANT: Réponds UNIQUEMENT avec un JSON valide sans backticks ni formatage markdown. | |
| Format exact attendu: | |
| {{ | |
| "questions": [ | |
| {{ | |
| "question": "La question", | |
| "options": ["Option A", "Option B", "Option C"], | |
| "correct_answer": "La bonne réponse (qui doit être une des options)" | |
| }} | |
| ] | |
| }} | |
| """) | |
| # Générer les questions | |
| response = await self.llm.agenerate([ | |
| prompt.format_messages( | |
| context=context, | |
| theme=theme, | |
| num_questions=num_questions | |
| ) | |
| ]) | |
| # Parser le JSON et créer les objets Question | |
| response_text = response.generations[0][0].text | |
| cleaned_response = self.clean_json_response(response_text) | |
| try: | |
| response_json = json.loads(cleaned_response) | |
| except json.JSONDecodeError as e: | |
| logger.error(f"Réponse brute: {response_text}") | |
| logger.error(f"Réponse nettoyée: {cleaned_response}") | |
| raise Exception(f"Erreur de parsing JSON: {str(e)}") | |
| questions = [] | |
| for q in response_json["questions"]: | |
| questions.append(Question( | |
| question=q["question"], | |
| options=q["options"], | |
| correct_answer=q["correct_answer"] | |
| )) | |
| return questions | |
| except Exception as e: | |
| print(f"Erreur dans generate_quiz: {str(e)}") | |
| raise e |