Update app.py

app.py

@@ -164,35 +164,35 @@ async def chat(request: ChatRequest):
     """Send a question to the chatbot with user session isolation"""
     if base_chatbot is None:
         raise HTTPException(status_code=503, detail="Chatbot not initialized")
-
+    
     if not request.question.strip():
         raise HTTPException(status_code=400, detail="Question cannot be empty")
-
+    
     if not request.user_id:
         raise HTTPException(status_code=400, detail="user_id is required")
-
+    
     try:
         logger.info(f"User {request.user_id}: {request.question[:50]}...")
-
+        
         # Get user session
         session = get_or_create_session(request.user_id)
-
+        
         # Resolve pronouns using user's context
         resolved_question = base_chatbot._resolve_pronouns_for_session(
-            request.question,
+            request.question, 
             session.conversation_context
         )
-
+        
         # Retrieve relevant chunks
         retrieved_data = base_chatbot._retrieve(resolved_question, k=20)
-
+        
         # Search user's chat history (not global)
         relevant_past_chats = base_chatbot._search_session_history(
             resolved_question,
             session.chat_history,
             k=5
         )
-
+        
         # Build prompt with user's context
         prompt = base_chatbot._build_prompt_for_session(
             resolved_question,
@@ -201,26 +201,23 @@ async def chat(request: ChatRequest):
             session.chat_history,
             session.conversation_context
         )
-
-        # Generate response
-        messages = [{"role": "user", "content": prompt}]
-
-        response = base_chatbot.llm_client.chat.completions.create(
+        
+        # ✅ CORRECT: Use text_generation for InferenceClient
+        answer = base_chatbot.llm_client.text_generation(
+            prompt,
             model="meta-llama/Llama-3.1-8B-Instruct",
-            messages=messages,
-            max_tokens=512,
-            temperature=0.3
+            max_new_tokens=512,
+            temperature=0.3,
+            return_full_text=False
         )
-
-        answer = response.choices[0].message.content
-
+        
         # Update user's conversation context
         base_chatbot._update_conversation_context_for_session(
             request.question,
             answer,
             session.conversation_context
         )
-
+        
         # Store in user's history
         chat_entry = {
             'timestamp': datetime.now().isoformat(),
@@ -229,7 +226,7 @@ async def chat(request: ChatRequest):
             'used_past_context': len(relevant_past_chats) > 0
         }
         session.chat_history.append(chat_entry)
-
+        
         response_data = ChatResponse(
             question=request.question,
             answer=answer,
@@ -240,15 +237,14 @@ async def chat(request: ChatRequest):
                 'current_context': session.conversation_context.get('current_employee')
             }
         )
-
+        
         logger.info(f"User {request.user_id}: Question processed successfully")
         return response_data
-
+        
     except Exception as e:
         logger.error(f"Error for user {request.user_id}: {e}")
         raise HTTPException(status_code=500, detail=f"Error: {str(e)}")
 
-
 @app.post("/api/reset")
 async def reset_chat(user_id: str):
     """Reset chat history for specific user"""