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

[CI] update typos config for CI pre-commit and fix some spells (#20919)

Browse Source 
Signed-off-by: Peter Pan <Peter.Pan@daocloud.io>
This commit is contained in:
Peter Pan
2025-07-16 12:12:40 +08:00
committed by GitHub
parent 6ebf313790
commit 1eb2b9c102
19 changed files with 200 additions and 196 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/model_executor/layers/fused_moe/fused_moe.py
View File

@@ -1172,7 +1172,7 @@ def fused_experts(
allow_cutlass_block_scaled_grouped_gemm: bool = False) -> torch.Tensor:
# For now, disable DeepGemm for small N (<= 512) until better
# permute/unpermute ops are available.
# However, on B200, we use DeepGemm for all cases becuase they only support
# However, on B200, we use DeepGemm for all cases because they only support
# E8M0 scale, which means we requantize the weight and input to the specific
# scale. Fallen back to cutlass or triton for some cases would cause
# accuracy issue.