xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-28 16:07:08 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bug] Fix modular_kernel: ZeroDivisionError: integer division or modulo by zero (#26528)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye 2025-10-09 18:13:27 -04:00 committed by GitHub
parent 6e783bc54b
commit 1ee35382cb
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 7 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/model_executor/layers/fused_moe/modular_kernel.py
View File

@ -717,10 +717,13 @@ class FusedMoEModularKernel(torch.nn.Module):
get num_chunks == 1. Take max(M, 1) to avoid divide by zero. get num_chunks == 1. Take max(M, 1) to avoid divide by zero.
If there are no tokens to process, the number of chunks will be zero. If there are no tokens to process, the number of chunks will be zero.
""" """
CHUNK_SIZE = ( CHUNK_SIZE = max(
max(M, 1) 1,
if not self.fused_experts.supports_chunking() (
else min(M, envs.VLLM_FUSED_MOE_CHUNK_SIZE) M
if not self.fused_experts.supports_chunking()
else min(M, envs.VLLM_FUSED_MOE_CHUNK_SIZE)
),
) )
num_chunks = cdiv(M, CHUNK_SIZE) num_chunks = cdiv(M, CHUNK_SIZE)
# If there are no tokens, then there should be no loop iterations. # If there are no tokens, then there should be no loop iterations.