xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-26 09:24:25 +08:00
Code Issues Packages Projects Releases Wiki Activity

stash

Browse Source 
Signed-off-by: Robert Shaw <robshaw@redhat.com>
This commit is contained in:
Robert Shaw 2025-07-20 22:33:17 +00:00
parent 91608889a4
commit be03d841f6
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/v1/engine/core_client.py
View File

@ -90,7 +90,8 @@ class EngineCoreClient(ABC):
client_args = (vllm_config, executor_class, log_stats, client_args = (vllm_config, executor_class, log_stats,
client_addresses, client_index) client_addresses, client_index)
if parallel_config.data_parallel_size > 1: if parallel_config.data_parallel_size > 1:
if parallel_config.data_parallel_external_lb: # if parallel_config.data_parallel_external_lb:
if False:
# External load balancer - client per DP rank. # External load balancer - client per DP rank.
return DPAsyncMPClient(*client_args) return DPAsyncMPClient(*client_args)
# Internal load balancer - client balances to all DP ranks. # Internal load balancer - client balances to all DP ranks.
@ -1023,6 +1024,7 @@ class DPLBAsyncMPClient(DPAsyncMPClient):
self, request: EngineCoreRequest) -> EngineIdentity: self, request: EngineCoreRequest) -> EngineIdentity:
# Engines are in rank order. # Engines are in rank order.
if (eng_index := request.data_parallel_rank) is None: if (eng_index := request.data_parallel_rank) is None:
# logger.info(f"{self.lb_engines=} | {self.core_engines=}")
if not self.lb_engines: if not self.lb_engines:
return self.core_engine return self.core_engine
# TODO use P2C alg for larger DP sizes # TODO use P2C alg for larger DP sizes