Skip to content

Update flash_attn_cute version to 0.0.2 and cute version to 4.4.2 - #13

Merged
demonatic merged 6 commits into
magi_attn_blackwell_supportfrom
update_cute_442
May 18, 2026
Merged

Update flash_attn_cute version to 0.0.2 and cute version to 4.4.2#13
demonatic merged 6 commits into
magi_attn_blackwell_supportfrom
update_cute_442

Conversation

@Strivin0311

@Strivin0311 Strivin0311 commented May 18, 2026

Copy link
Copy Markdown
Collaborator

DONE

  • updated flash_attn_cute version to 0.0.2.
  • updated cute version to 4.4.2 in requirements.txt.
  • cherry-picked commits from magi_attn_blackwell_support_lts branch, including d728c00, d728c00 and d9c568d.
  • fixed pre-commit.

@Strivin0311 Strivin0311 changed the title Update cute version to 4.4.2 Update flash_attn_cute version to 0.0.2 and cute version to 4.4.2 May 18, 2026
Strivin0311 and others added 4 commits May 18, 2026 11:03
…mpatibility

cutlass-dsl 4.4.2 moved TCGen05Mma support into sm100.py's overridden
_make_sync_object. Calling PipelineAsync._make_sync_object (sm90.py)
directly hits an assert because it doesn't recognize TCGen05Mma.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…for cutlass-dsl 4.4.2

cutlass-dsl 4.4.2 passes loc/ip kwargs through @dsl_user_op decorated
methods. The overridden producer_acquire and consumer_release in both
PipelineTmaAsync and PipelineTmaUmma lacked these parameters, causing
DSLRuntimeError when producer_tail calls producer_acquire(state, loc=loc).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
auto_user[Jager(Jager)] and others added 2 commits May 18, 2026 15:08
@demonatic
demonatic merged commit ee1d151 into magi_attn_blackwell_support May 18, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants