xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-13 18:25:40 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BugFix][DeepSeek-V3.2] Fix backend selection logic for Blackwell (#30195)

Browse Source
This commit is contained in:
Lucas Wilkinson 2025-12-07 10:53:51 -05:00 committed by GitHub
parent b952f4d3c3
commit 0044c4038c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/platforms/cuda.py
View File

@ -182,8 +182,8 @@ class CudaPlatformBase(Platform):
if vllm_config.attention_config.backend is None: if vllm_config.attention_config.backend is None:
# Default case # Default case
if cls.is_device_capability(100): if cls.is_device_capability(100) and not use_sparse:
# Blackwell => Force CutlassMLA. # Blackwell => Force CutlassMLA (unless sparse, i.e. DSv3.2).
use_cutlass_mla = True use_cutlass_mla = True
# Set the backend in AttentionConfig so it's used during # Set the backend in AttentionConfig so it's used during
# backend selection # backend selection