Disable Cascade Attention for Batch Invariance (#32561)

Signed-off-by: frankwang28 <frank.wbb@hotmail.com>
Signed-off-by: Frank Wang <41319051+frankwang28@users.noreply.github.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
This commit is contained in:
Frank Wang
2026-01-30 10:00:46 -05:00
committed by GitHub
co-authored by Wentao Ye
parent ae5b7aff2b
commit 8f5d51203b
6 changed files with 60 additions and 9 deletions
@@ -1005,7 +1005,9 @@ def override_envs_for_invariance(
):
supported_backends = [
AttentionBackendEnum.FLASH_ATTN, # best supported backend
AttentionBackendEnum.FLASHINFER,
# FlashInfer temporarily disabled due to invariant CTA sizes.
# See FlashInfer issue #2424
# AttentionBackendEnum.FLASHINFER,
AttentionBackendEnum.FLASH_ATTN_MLA,
AttentionBackendEnum.TRITON_MLA,
# Not yet supported MLA backends