EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Fix error due to CustomAllreduce setup failure (#4815)

Browse Source 
Signed-off-by: Kebe <mail@kebe7jun.com>
This commit is contained in:
Kebe
2025-03-28 09:52:10 +08:00
committed by GitHub
parent bb0fd749a6
commit 10a9ab7b07
1 changed files with 10 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
python/sglang/srt/distributed/parallel_state.py
View File

@@ -264,10 +264,16 @@ class GroupCoordinator:
self.ca_comm: Optional[CustomAllreduce] = None self.ca_comm: Optional[CustomAllreduce] = None
if use_custom_allreduce and self.world_size > 1: if use_custom_allreduce and self.world_size > 1:
# Initialize a custom fast all-reduce implementation. # Initialize a custom fast all-reduce implementation.
self.ca_comm = CustomAllreduce( try:
group=self.cpu_group, self.ca_comm = CustomAllreduce(
device=self.device, group=self.cpu_group,
) device=self.device,
)
except Exception as e:
logger.warning(
f"Setup Custom allreduce failed with {e}. To silence this "
"warning, specify --disable-custom-all-reduce explicitly."
)
from sglang.srt.distributed.device_communicators.hpu_communicator import ( from sglang.srt.distributed.device_communicators.hpu_communicator import (
HpuCommunicator, HpuCommunicator,