Fix multi-tokenizer batch request output routing (health stuck at 503) (#26831)
This commit is contained in:
@@ -562,11 +562,11 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
|
|||||||
f"routed_dp_rank={obj.routed_dp_rank} out of range [0, {dp_size})"
|
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)
|
self._init_req_state(obj, request)
|
||||||
if self.server_args.language_only:
|
if self.server_args.language_only:
|
||||||
self._handle_epd_disaggregation_encode_request(obj)
|
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
|
# Log the request
|
||||||
self.request_logger.log_received_request(obj, self.tokenizer, request)
|
self.request_logger.log_received_request(obj, self.tokenizer, request)
|
||||||
|
|||||||
Reference in New Issue
Block a user