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

[CPU] Fix fallback allgather issue (#8041)

Browse Source
This commit is contained in:
blzheng
2025-08-07 15:08:18 +08:00
committed by GitHub
parent b7cd743038
commit 62f8eb48b1
1 changed files with 12 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -650,17 +650,19 @@ class GroupCoordinator:
output_size, dtype=input_.dtype, device=input_.device output_size, dtype=input_.dtype, device=input_.device
) )
if input_.is_cpu: # All-gather.
if is_shm_available(input_.dtype, self.world_size, self.local_size): if input_.is_cpu and is_shm_available(
input_.dtype, self.world_size, self.local_size
):
return torch.ops.sgl_kernel.shm_allgather(input_, dim) return torch.ops.sgl_kernel.shm_allgather(input_, dim)
else:
if input_.is_cpu:
torch.distributed.all_gather_into_tensor( torch.distributed.all_gather_into_tensor(
output_tensor, input_, group=self.device_group output_tensor, input_, group=self.device_group
) )
return output_tensor else:
# All-gather.
self.all_gather_into_tensor(output_tensor, input_) self.all_gather_into_tensor(output_tensor, input_)
# Reshape # Reshape
output_tensor = output_tensor.reshape((world_size,) + input_size) output_tensor = output_tensor.reshape((world_size,) + input_size)
output_tensor = output_tensor.movedim(0, dim) output_tensor = output_tensor.movedim(0, dim)