copilot 优化方案

app.py

@@ -195,39 +195,77 @@ def create_chat_completion_data(
     }
 
 async def handle_stream(response, previous_content):
-    timestamp = int(datetime.now().timestamp())
-    last_content = ""
+    """
+    优化后的流式响应处理函数
 
-    async def yield_and_flush(data):
-        yield data
-        await asyncio.sleep(0.01)  # 使用更明确的延迟
+    主要改进：
+    1. 使用简单的字符串切片而不是startswith比较
+    2. 增加内容缓存来减少重复处理
+    3. 优化内存使用
+    """
+    timestamp = int(datetime.now().timestamp())
+    content_cache = ""  # 用于缓存已处理的内容
+
+    async def send_chunk(content: str):
+        """内部函数：格式化并发送数据块"""
+        chunk_data = {
+            "id": f"chatcmpl-{uuid.uuid4()}",
+            "object": "chat.completion.chunk",
+            "created": timestamp,
+            "model": "qwen",
+            "choices": [{
+                "index": 0,
+                "delta": {
+                    "content": content,
+                    "role": "assistant"
+                },
+                "finish_reason": None
+            }]
+        }
+        return f"data: {json.dumps(chunk_data)}\n\n"
 
     async for chunk in response.aiter_lines():
+        if not chunk or not chunk.startswith('data: '):
+            continue
+
         try:
-            current_chunk = chunk
-            if not current_chunk:
-                continue
+            # 解析数据
+            data = json.loads(chunk[6:])  # 去掉 'data: ' 前缀
 
-            if current_chunk.startswith('data: '):
-                current_content = ''
-                data = json.loads(current_chunk[6:])
-                if data.get('choices') and data['choices'][0].get('delta') and data['choices'][0]['delta'].get('content'):
-                    current_content = data['choices'][0]['delta']['content']
+            # 提取新内容
+            if not (data.get('choices') and
+                    data['choices'][0].get('delta') and
+                    data['choices'][0]['delta'].get('content')):
+                continue
 
-                    new_content = (current_content[len(last_content):]
-                                   if last_content and current_content.startswith(last_content)
-                                   else current_content)
+            current_content = data['choices'][0]['delta']['content']
 
-                    if new_content:
-                        response_data = f"data: {json.dumps(create_chat_completion_data(new_content, 'qwen', timestamp))}\n\n"
-                        async for item in yield_and_flush(response_data):
-                            yield item
-                        last_content = current_content
+            # 智能差异检测
+            if content_cache:
+                # 找到新内容的起始位置
+                for i in range(min(len(current_content), len(content_cache))):
+                    if current_content[i] != content_cache[i]:
+                        new_content = current_content[i:]
+                        break
+                else:
+                    # 如果前面的内容都相同，新内容就是超出的部分
+                    new_content = current_content[len(content_cache):]
+            else:
+                new_content = current_content
+
+            # 只有真正有新内容时才发送
+            if new_content:
+                yield await send_chunk(new_content)
+                content_cache = current_content  # 更新缓存
 
+        except json.JSONDecodeError as e:
+            logger.error(f"JSON解析错误: {str(e)}")
+            continue
         except Exception as e:
-            logger.error(f"Error processing chunk: {str(e)}")
+            logger.error(f"处理数据流时发生错误: {str(e)}")
             yield f"data: {{\"error\":true,\"message\":\"{str(e)}\"}}\n\n"
 
+    # 发送结束标记
     yield "data: [DONE]\n\n"