Update app.py

app.py

@@ -108,7 +108,7 @@ def respond(message, history):
         llm = HuggingFaceInferenceAPI(
             model_name=selected_llm_model_name,
             contextWindow=8192,  # Context window size (typically max length of the model)
-            maxTokens=1024,  # Tokens per response generation (512-1024 works well for detailed answers)
+            maxTokens=2048,  # Tokens per response generation (512-1024 works well for detailed answers)
             temperature=0.3,  # Lower temperature for more focused answers (0.2-0.4 for factual info)
             topP=0.9,  # Top-p sampling to control diversity while retaining quality
             frequencyPenalty=0.5,  # Slight penalty to avoid repetition
@@ -120,8 +120,8 @@ def respond(message, history):
         query_engine = vector_index.as_query_engine(llm=llm)
         bot_message = query_engine.query(message)
 
-        print(f"\n{datetime.now()}:{selected_llm_model_name}:: {message} --> {str(bot_message)}\n")
-        return f"{selected_llm_model_name}:\n{str(bot_message)}"
+        print(f"\n{datetime.now()}:{selected_llm_model_name} :: {message} --> {str(bot_message)}\n")
+        return f"{selected_llm_model_name}:\n\n{str(bot_message)}"
     except Exception as e:
         if str(e) == "'NoneType' object has no attribute 'as_query_engine'":
             return "Please upload a file."