Update app.py

app.py

@@ -86,22 +86,6 @@ class ChatRequest(BaseModel):
     top_p: float = 0.95
     temperature: float = 0.7
 
-@spaces.GPU(duration=0)
-def generate_chat_response(request, model_data):
-    try:
-        user_input = normalize_input(request.message)
-        llm = model_data['model']
-        response = llm.create_chat_completion(
-            messages=[{"role": "user", "content": user_input}],
-            top_k=request.top_k,
-            top_p=request.top_p,
-            temperature=request.temperature
-        )
-        reply = response['choices'][0]['message']['content']
-        return {"response": reply, "literal": user_input, "model_name": model_data['name']}
-    except Exception:
-        pass
-
 def normalize_input(input_text):
     return input_text.strip()
 
@@ -128,6 +112,16 @@ def remove_repetitive_responses(responses):
             unique_responses.append(response)
     return unique_responses
 
+@spaces.GPU(duration=0)
+def generate_chat_response(request, model_data):
+    try:
+        user_input = normalize_input(request.message)
+        llm = model_data['model']
+        response = llm(user_input, top_k=request.top_k, top_p=request.top_p, temperature=request.temperature)
+        return {"model": model_data['name'], "response": response}
+    except Exception:
+        pass
+
 @spaces.GPU(duration=0)
 async def generate(request: ChatRequest):
     try:
@@ -146,7 +140,7 @@ async def generate(request: ChatRequest):
             raise HTTPException(status_code=500, detail="Error: No responses generated.")
         
         responses = remove_repetitive_responses(responses)
-        best_response = select_best_response(responses)
+        best_response = responses[0] if responses else {}
         return {
             "best_response": best_response,
             "all_responses": responses
@@ -154,8 +148,5 @@ async def generate(request: ChatRequest):
     except Exception:
         pass
 
-def select_best_response(responses):
-    return responses[0] if responses else {}
-
 if __name__ == "__main__":
     uvicorn.run(app, host="0.0.0.0", port=7860)