Skip to content

Support H-max to 256 in JAX/SPDA for Blackwell. #33504

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
+2 −2
Open

Support H-max to 256 in JAX/SPDA for Blackwell. #33504

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
86d3c6d
Support H-max to 256 in JAX/SPDA for Blackwell.
mingxu1067 Nov 24, 2025
3feb474
Apply check_compute_capability
mingxu1067 Nov 24, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions jax/_src/cudnn/fused_attention_stablehlo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -379,8 +379,8 @@ def check_is_flash_attention(
else:
# bf16/fp16 attention conditions
# Check the head dim.
is_on_hopper = is_cuda_compute_capability_equal("9.0")
H_max = 256 if is_on_hopper else 128
is_hopper_or_later = check_compute_capability("9.0")
H_max = 256 if is_hopper_or_later else 128
# check if multi-head latent attention is needed
is_mla = qH != vH
if not (qH <= H_max and qH % 8 == 0):
Expand Down