再次修改回复格式的问题

app.py

@@ -30,34 +30,18 @@ llm = Llama(
 print("Model loaded successfully!")
 
 def chat(message, history):
-    """Handle chat interactions"""
+    llm_message = message["content"]
 
     conversation = "System: You are a helpful assistant.\n"
 
-    # Only use last 3 turns
-    recent_history = history[-3:] if len(history) > 3 else history
-
-    for msg in recent_history:
-        # ChatInterface passes dict messages
-        if isinstance(msg, dict):
-            role = msg.get("role")
-            content = msg.get("content", "")
-
-            if role == "user":
-                conversation += f"User: {content}\n"
-            elif role == "assistant":
-                conversation += f"Assistant: {content}\n"
-
-        # fallback: old tuple/list history
+    for msg in history[-3:]:
+        if msg["role"] == "user":
+            conversation += f"User: {msg['content']}\n"
         else:
-            human, assistant = msg
-            conversation += f"User: {human}\n"
-            conversation += f"Assistant: {assistant or ''}\n"
+            conversation += f"Assistant: {msg['content']}\n"
 
-    # Add current user message
-    conversation += f"User: {message}\nAssistant:"
+    conversation += f"User: {llm_message}\nAssistant:"
 
-    # LLM inference
     response = llm(
         conversation,
         max_tokens=128,
@@ -65,11 +49,12 @@ def chat(message, history):
         top_p=0.9,
         top_k=40,
         repeat_penalty=1.1,
-        stop=["User:", "Assistant:"],
-        echo=False
+        stop=["User:", "Assistant:"]
     )
 
-    return response["choices"][0]["text"].strip()
+    reply = response["choices"][0]["text"].strip()
+
+    return {"role": "assistant", "content": reply}
 
 
 # Create interface WITHOUT example caching