A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 12, 17:15 UTC
[tiny] Remove unsupported TRITON_MLA backend from batch invariance (#28832)
Signed-off-by: Bram Wasti <bwasti@meta.com> Signed-off-by: Bram Wasti <bwasti@fb.com> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
Bram Wasti committed
Nov 22, 2025 at 08:00 UTC
5f7209a793ec553889f8ba9972a0034393a6b196
1 file changed
+1
-1
vllm/model_executor/layers/batch_invariant.py
+1
-1
@@ -805,11 +805,11 @@ def override_envs_for_invariance():
805
"FLASH_ATTN", # best supported backend
806
"FLASHINFER",
807
"FLASH_ATTN_MLA",
808
- "TRITON_MLA",
808
# Not yet supported MLA backends
809
# "FLASHMLA",
810
# "FLEX_ATTENTION", # IMA issue even if we disable batch invariance
811
# "FLASHINFER_MLA", https://github.com/vllm-project/vllm/pull/28967
812
+ # "TRITON_MLA",
813
]
814
if curr_attn_backend not in supported_backends:
815
warning = (