Do not output response stream

main.py

@@ -72,6 +72,9 @@ async def error_handling_wrapper(generator, status_code=200):
                 first_item_str = first_item_str[5:]
             first_item_str = json.loads(first_item_str)
         if isinstance(first_item_str, dict) and 'error' in first_item_str:
+            print('\033[31m')
+            print(f"first_item_str: {first_item_str}")
+            print('\033[0m')
             # 如果第一个 yield 的项是错误信息，抛出 HTTPException
             raise HTTPException(status_code=status_code, detail=first_item_str)
 

response.py

@@ -42,7 +42,7 @@ async def fetch_gemini_response_stream(client, url, headers, payload, model):
             buffer += chunk
             while "\n" in buffer:
                 line, buffer = buffer.split("\n", 1)
-                print(line)
+                # print(line)
                 if line and '\"text\": \"' in line:
                     try:
                         json_data = json.loads( "{" + line + "}")
@@ -82,7 +82,7 @@ async def fetch_gpt_response_stream(client, url, headers, payload):
             buffer += chunk.decode('utf-8')
             while "\n" in buffer:
                 line, buffer = buffer.split("\n", 1)
-                print(line)
+                # print(line)
                 yield line + "\n"
 
 async def fetch_claude_response_stream(client, url, headers, payload, model):
@@ -102,7 +102,7 @@ async def fetch_claude_response_stream(client, url, headers, payload, model):
             buffer += chunk.decode('utf-8')
             while "\n" in buffer:
                 line, buffer = buffer.split("\n", 1)
-                print(line)
+                # print(line)
 
                 if line.startswith("data:"):
                     print(line)