diff --git a/python/sglang/srt/entrypoints/http_server.py b/python/sglang/srt/entrypoints/http_server.py index ac38b4668..a2f48d9b7 100644 --- a/python/sglang/srt/entrypoints/http_server.py +++ b/python/sglang/srt/entrypoints/http_server.py @@ -2041,8 +2041,10 @@ async def vertex_generate( **(vertex_req.parameters or {}), ) ret = await generate_request(req, raw_request) - if isinstance(ret, Response): - return ret + if isinstance(ret, Response) and ret.status_code == 200 and hasattr(ret, "body"): + return ORJSONResponse({"predictions": orjson.loads(ret.body)}) + return ret + return ORJSONResponse({"predictions": ret})