Update app.py

app.py

@@ -76,17 +76,21 @@ def chat_completions_proxy():
                     event_type = gpt_oss_data.get('type')
                     update_type = gpt_oss_data.get('update', {}).get('type')
                     
+                    # 关键逻辑: 每收到一条思考，就立即格式化并发送
                     if event_type == 'thread.item_updated' and update_type == 'cot.entry_added':
                         thought = gpt_oss_data['update']['entry']['content']
-                        formatted_thought = f"[思考中] {thought}"
+                        # 将单条思考包装在它自己的Markdown代码块中
+                        formatted_thought_block = f"```markdown\n[思考中] {thought}\n```"
                         
-                        # 为了兼容性，不逐字发送，而是一次性发送整条思考
-                        thought_chunk = create_openai_chunk(formatted_thought)
+                        # 1. 发送包含代码块的chunk
+                        thought_chunk = create_openai_chunk(formatted_thought_block)
                         yield f"data: {json.dumps(thought_chunk)}\n\n"
-                        # 发送换行符以分隔
+                        
+                        # 2. 发送一个包含换行符的chunk，以在视觉上分隔不同的思考块
                         newline_chunk = create_openai_chunk("\n\n")
                         yield f"data: {json.dumps(newline_chunk)}\n\n"
-
+                    
+                    # 正常流式传输最终答案
                     if event_type == 'thread.item_updated' and update_type == 'assistant_message.content_part.text_delta':
                         delta_content = gpt_oss_data['update'].get('delta', '')
                         text_chunk = create_openai_chunk(delta_content)
@@ -98,7 +102,7 @@ def chat_completions_proxy():
 
         return Response(stream_formatter(), mimetype='text/event-stream')
     
-    else:
+    else: # 非流式请求逻辑保持不变
         try:
             full_response_content = ""
             for gpt_oss_data in _internal_proxy_stream():