A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 3, 16:30 UTC
[V1][DP][LB] Publish request counts at the start of each engine step (#41626)
Signed-off-by: Vadim Gimpelson <vadim.gimpelson@gmail.com> Signed-off-by: Vadim Gimpelson <156319763+vadiklyutiy@users.noreply.github.com> Co-authored-by: Nick Hill <nickhill123@gmail.com>
Vadim Gimpelson committed
May 14, 2026 at 19:39 UTC
f60c6b33a5c637315624c841775b5efdf6fb46d6
1 file changed
+2
vllm/v1/engine/core.py
+2
@@ -1817,6 +1817,8 @@ class DPEngineCoreProc(EngineCoreProc):
1817
while self._handle_shutdown():
1818
# 1) Poll the input queue until there is work to do.
1819
self._process_input_queue()
1820
+ # Publish request counts before and after GPU step to ensure freshness.
1821
+ self._maybe_publish_request_counts()
1822
1823
if self.eep_scaling_state is not None:
1824
_ = self.eep_scaling_state.progress()