Update app/app.py

app/app.py

@@ -77,7 +77,7 @@ logger.info(f"[INFO] LLM_TIMEOUT_SECONDS set to: {LLM_TIMEOUT_SECONDS} seconds."
 
 async def generate_llm_response(prompt: str):
     """Helper function to run synchronous LLM inference."""
-    response = llm(prompt, max_tokens=512, stop=["###"], temperature=0.2, echo=False)
+    response = llm(prompt, max_tokens=1024, stop=["###"], temperature=0.2, echo=False)
     answer = response["choices"][0]["text"].strip()
     if not answer:
         raise ValueError("Empty response from LLM")