iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[XPU] fix all_reduce all-zero accuracy issue under torch.compile (#39844)

Browse Source 
Signed-off-by: Chaojun Zhang <chaojun.zhang@intel.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Chaojun Zhang
2026-04-18 02:33:07 +00:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 48a65ccb02
commit 993859ceb0
1 changed files with 4 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/distributed/device_communicators/xpu_communicator.py
+4 -3
View File
@@ -47,9 +47,10 @@ class XpuCommunicator(DeviceCommunicatorBase):
self.all2all_manager = AgRsAll2AllManager(self.cpu_group)
logger.info("Using AgRs manager on XPU device.")
def all_reduce(self, input_) -> torch.Tensor:
dist.all_reduce(input_, group=self.device_group)
return input_
def all_reduce(self, input_: torch.Tensor) -> torch.Tensor:
output = input_.clone() if torch.compiler.is_compiling() else input_
dist.all_reduce(output, group=self.device_group)
return output
def reduce_scatter(self, input_: torch.Tensor, dim: int = -1):
world_size = self.world_size