minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Diffusion][CPU] Enable MiniMax-H3 on Xeon CPU (#35147)

Browse Source
This commit is contained in:
Xinguo Zhu
2026-09-08 10:56:42 +08:00
committed by GitHub
parent 73c4cdb795
commit 2bf04f3a67
2 changed files with 6 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/multimodal_gen/runtime/distributed/group_coordinator.py
+4 -1
View File
@@ -441,7 +441,10 @@ class GroupCoordinator:
if current_platform.is_cpu() and is_shm_available( if current_platform.is_cpu() and is_shm_available(
input_.dtype, self.world_size, len(self.ranks) input_.dtype, self.world_size, len(self.ranks)
): ):
return torch.ops.sgl_kernel.shm_allgather(input_, dim) output_tensor = torch.ops.sgl_kernel.shm_allgather(input_, dim)
if separate_tensors:
return list(output_tensor.chunk(world_size, dim=dim))
return output_tensor
else: else:
all_gather_single(output_tensor, input_, group=self.device_group) all_gather_single(output_tensor, input_, group=self.device_group)
python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/minimax_h3/stages/denoising.py
+2 -1
View File
@@ -676,11 +676,12 @@ class MiniMaxH3DenoisingStage(DenoisingStage):
if not ( if not (
current_platform.is_cuda() current_platform.is_cuda()
or current_platform.is_cpu()
or current_platform.is_mps() or current_platform.is_mps()
or current_platform.is_npu() or current_platform.is_npu()
): ):
raise RuntimeError( raise RuntimeError(
"MiniMax H3 full-loop denoise requires CUDA, MPS, or Ascend NPU" "MiniMax H3 full-loop denoise requires CPU, CUDA, MPS, or Ascend NPU"
) )
device = current_platform.get_local_torch_device() device = current_platform.get_local_torch_device()