xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-27 07:07:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BugFix] Fix MLA + V1 + TP==1 causing reinitialization of cuda context (#14910)

Browse Source
This commit is contained in:
Lucas Wilkinson 2025-03-16 23:35:37 -04:00 committed by GitHub
parent 7f6c5ee06c
commit 1e799b7ec1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/platforms/cuda.py
View File

@ -152,7 +152,7 @@ class CudaPlatformBase(Platform):
# here # here
use_flashmla = (envs.VLLM_ATTENTION_BACKEND is None \ use_flashmla = (envs.VLLM_ATTENTION_BACKEND is None \
or envs.VLLM_ATTENTION_BACKEND == "FLASHMLA") or envs.VLLM_ATTENTION_BACKEND == "FLASHMLA")
from vllm.attention.backends.flashmla import is_flashmla_supported from vllm.attention.ops.flashmla import is_flashmla_supported
if use_flashmla and is_flashmla_supported()[0] \ if use_flashmla and is_flashmla_supported()[0] \
and cache_config.block_size != 64: and cache_config.block_size != 64:
cache_config.block_size = 64 cache_config.block_size = 64