Skip to content

Let splash attention exceed 524,288 tokens when packing is off #4929

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

Let splash attention exceed 524,288 tokens when packing is off #4929

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
43f4bce
Let splash attention exceed 524,288 tokens when packing is off
WandLZhang Aug 18, 2026
83f18aa
Drop the segment ids before the early returns
WandLZhang Aug 18, 2026
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
9 changes: 9 additions & 0 deletions src/maxtext/layers/attention_op.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1986,6 +1986,15 @@ def wrap_flash_attention(
sinks,
indexer_mask,
):
# Splash prefetches the segment ids into SMEM, and SMEM is 1 MB per core.
# This limits the segmented kernel to approximately 524,288 tokens. When
# packing is off there is one segment per example, so the segment ids are
# constant and the non-segmented kernel gives the same result. Drop the
# ids here, before any early return, so every path benefits.
if not self.config.packing:
decoder_segment_ids_q = None
decoder_segment_ids_kv = None

if use_tokamax_ring:
attention_output = tokamax_ring_attention.call_ring_attention(
query,
Expand Down