biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Kernels] Use empty for modular MoE workspaces (#19667)

Browse Source 
Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2025-06-16 10:58:01 -04:00
committed by GitHub
parent 836d4ce140
commit 5e5baa91aa
2 changed files with 5 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/layers/fused_moe/modular_kernel.py
View File

@@ -426,10 +426,10 @@ class FusedMoEModularKernel(torch.nn.Module):
# We can reuse the memory between cache1 and cache3 because by the
# time we need cache3, we're done with cache1.
workspace13 = torch.zeros(prod(workspace13_shape),
workspace13 = torch.empty(prod(workspace13_shape),
device=a1.device,
dtype=workspace_dtype)
workspace2 = torch.zeros(prod(workspace2_shape),
workspace2 = torch.empty(prod(workspace2_shape),
device=a1.device,
dtype=workspace_dtype)