Update app.py

app.py

@@ -64,9 +64,14 @@ def format_message(content):
 
 
 def user_input(user_message, history):
-    return "", history  # 不再主动添加消息到历史记录
+    """处理用户输入"""
+    if history is None:
+        history = []
+    history = history + [[user_message, None]]  # 只添加用户消息
+    return "", history
 
 def predict(message, chat_history, system_msg, temperature, top_p, repetition_penalty):
+    """处理AI回复"""
     headers = {
         "Content-Type": "application/json",
         "Authorization": f"Bearer {API_KEY}"
@@ -74,10 +79,13 @@ def predict(message, chat_history, system_msg, temperature, top_p, repetition_pe
     
     if chat_history is None:
         chat_history = []
+        
+    # 不需要再添加用户消息,因为已经在user_input中添加了
+    history = chat_history.copy()
     
     # 构造对话历史
     messages = [{"role": "system", "content": system_msg}]
-    for msg in chat_history:
+    for msg in history:
         messages.append({"role": "user", "content": msg[0]})
         if msg[1]:
             messages.append({"role": "assistant", "content": msg[1]})
@@ -99,23 +107,21 @@ def predict(message, chat_history, system_msg, temperature, top_p, repetition_pe
     )
     
     buffer = ""
-    history = chat_history.copy() + [[message, None]]  # 初始化新对话
+    first_chunk = True
     
-    # 流式处理关键修改点
+    # 流式处理
     for chunk in response.iter_lines():
         if chunk:
-            chunk_str = chunk.decode("utf-8").replace("data: ", "")
-            try:
+            chunk_str = chunk.decode("utf-8").:
                 chunk_data = json.loads(chunk_str)
                 if "choices" in chunk_data:
                     delta = chunk_data["choices"][0].get("delta", {})
                     if "content" in delta:
-                        # 增量更新缓冲区
                         buffer += delta["content"]
                         
-                        # 只更新当前消息的最新状态
+                        # 更新最后一条消息的回复
                         new_history = history.copy()
-                        new_history[-1][1] = format_message(buffer)  # 直接使用字符串
+                        new_history[-1][1] = format_message(buffer)
                         yield new_history
             except Exception as e:
                 print(f"Error in processing chunk: {e}")
@@ -129,6 +135,7 @@ def predict(message, chat_history, system_msg, temperature, top_p, repetition_pe
 
 
 
+
 # JavaScript to ensure the details tag is working properly
 js_script = """
 <script>