xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-09 12:25:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] [B200] cutlass_mla - ensure kv_split == 1 for batch size > 1 (#25509)

Browse Source 
Signed-off-by: Alexander Matveev <amatveev@redhat.com>
This commit is contained in:
Alexander Matveev 2025-09-23 19:57:55 -04:00 committed by GitHub
parent e0b24ea030
commit 1210e4d95b
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
csrc/attention/mla/cutlass_sm100_mla/device/sm100_mla.hpp
View File

@ -135,10 +135,10 @@ public:
max_splits = min(16, max_splits);
// TODO: This avoids a hang when the batch size larger than 1 and
// there is more than 4 kv_splits.
// there is more than 1 kv_splits.
// Discuss with NVIDIA how this can be fixed.
if (B > 1) {
max_splits = min(2, max_splits);
max_splits = min(1, max_splits);
}
// printf(" max_splits = %d\n", max_splits);