修复respond函数的参数处理逻辑

app.py

@@ -53,17 +53,13 @@ def respond(
     temperature: float,
     top_p: float,
 ):
-    print('message', message)
-    print('history', history)
     if len(history) == 0 or history[0]['role'] != 'system':
-        messages = [{"role": "system", "content": """You are Softie, or 小软 in Chinese.
+        history.insert(0, {"role": "system", "content": """You are Softie, or 小软 in Chinese.
 You are an intelligent assistant developed by the School of Software at Hefei University of Technology.
-You like to chat with people and help them solve problems."""}]
-    for m in history:
-        messages.append({"role": m["role"], "content": m["content"]})
-    messages.append({"role": "user", "content": message})
-    text_inputs = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
-    image_inputs, video_inputs = process_vision_info(messages)
+You like to chat with people and help them solve problems."""})
+    history.append({"role": "user", "content": message})
+    text_inputs = processor.apply_chat_template(history, tokenize=False, add_generation_prompt=True)
+    image_inputs, video_inputs = process_vision_info(history)
     for response in infer((text_inputs, image_inputs, video_inputs), max_tokens, temperature, top_p):
         yield response