From 1eadb7a173fdfc33f96e8cf019616a0abdc6722e Mon Sep 17 00:00:00 2001 From: ybyang <10629930+whybeyoung@users.noreply.github.com> Date: Sun, 31 May 2026 15:31:23 +0800 Subject: [PATCH] Fix multi-tokenizer batch request output routing (health stuck at 503) (#26831) --- python/sglang/srt/managers/tokenizer_manager.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/python/sglang/srt/managers/tokenizer_manager.py b/python/sglang/srt/managers/tokenizer_manager.py index 656bfc7b6..df9a25714 100644 --- a/python/sglang/srt/managers/tokenizer_manager.py +++ b/python/sglang/srt/managers/tokenizer_manager.py @@ -562,11 +562,11 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin): f"routed_dp_rank={obj.routed_dp_rank} out of range [0, {dp_size})" ) + if self.server_args.tokenizer_worker_num > 1: + self._attach_multi_http_worker_info(obj) self._init_req_state(obj, request) if self.server_args.language_only: self._handle_epd_disaggregation_encode_request(obj) - if self.server_args.tokenizer_worker_num > 1: - self._attach_multi_http_worker_info(obj) # Log the request self.request_logger.log_received_request(obj, self.tokenizer, request)