xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-01-23 18:34:28 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix][EPLB] Disabled shared expert overlap when EPLB is enabled (#28377)

Browse Source 
Signed-off-by: Sage Moore <sage@neuralmagic.com>
Signed-off-by: Sage Moore <sagemoore@utexas.edu>
Signed-off-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
This commit is contained in:
Sage Moore 2025-11-10 12:39:19 -08:00 committed by GitHub
parent 9c84ca8293
commit 40d33264c6
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 10 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
vllm/model_executor/layers/fused_moe/shared_fused_moe.py
View File

@ -28,13 +28,18 @@ class SharedFusedMoE(FusedMoE):
super().__init__(**kwargs)
self._shared_experts = shared_experts
# Disable shared expert overlap if we are not using
# flashinfer + DP since there is nothing to be gained in this case.
# Disabling the overlap optimization also prevents the shared experts
# from being hidden from torch.compile.
# Disable shared expert overlap if we are using eplb, because of
# correctness issues, or if using flashinfer with DP, since there
# is nothing to be gained in this case. Disabling the overlap
# optimization also prevents the shared experts from being hidden
# from torch.compile.
self.use_overlapped = (
use_overlapped
and not (self.use_flashinfer_cutlass_kernels and self.dp_size > 1)
and not (
# TODO(wentao): find the root cause and remove this condition
self.enable_eplb
or (self.use_flashinfer_cutlass_kernels and self.dp_size > 1)
)
and self._shared_experts is not None
)