Skip to content

Pull requests: Dao-AILab/flash-attention

New pull request New
115 Open 445 Closed
115 Open 445 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Add SM120 (Blackwell GeForce / DGX Spark) forward pass support
#2268 opened Feb 20, 2026 by blake-snc Loading…
2 of 4 tasks
[WIP,Cute,Flex,Sm100] vectorized mask mod application
#2261 opened Feb 17, 2026 by reubenconducts • Draft
Add sm_121 support for NVIDIA GB10 GPUs (CUDA 13.0+)
#2257 opened Feb 13, 2026 by askliar Loading…
1
5
Deterministic backward for blocksparse impl
#2253 opened Feb 12, 2026 by drisspg Loading…
5
correct max_seqlen_q for performance.
#2252 opened Feb 12, 2026 by lightzhan-intellif Loading…
4
[Cute] Handle window_size=(-1, -1) for non-local attention
#2251 opened Feb 11, 2026 by henrylhtsang Loading…
8
Add two-level accumulation for SM90 FP8 FWD to mitigate long-context degradation
#2250 opened Feb 11, 2026 by jmkuebler Loading…
Fix directory path in README instructions
#2249 opened Feb 11, 2026 by Chuge0335 Loading…
support the flash api for Ascend
#2246 opened Feb 10, 2026 by AnyFree813 Loading…
Add gfx1150/gfx1151 (RDNA 3.5) to RDNA_ARCHS
#2243 opened Feb 9, 2026 by rwfsmith Loading…
2
[AMD] Migrate to Triton Backend to Aiter
#2230 opened Feb 4, 2026 by micmelesse Loading…
1
Nicer headdim error message
#2227 opened Feb 4, 2026 by drisspg Loading…
8
[WIP] varlen blocksparsity
#2224 opened Feb 2, 2026 by reubenconducts • Draft
2
[Draft][Cute,Fwd,Sm120] FA Cute DSL sm12x
#2222 opened Feb 2, 2026 by johnnynunez • Draft
1
[Ai-assisted] CLC work stealing
#2218 opened Jan 31, 2026 by drisspg Loading…
11
[ROCM] Add support with Infinity Cache (LLC) awareness for performance improvement - [PR#2147 rebased on PR#2178]
#2217 opened Jan 29, 2026 by tianwyan Loading…
16
Add shift scheduler for deterministic full‑mask FA3 bwd on Hopper (sm90)
#2207 opened Jan 23, 2026 by tie-pilot-qxw Loading…
3
Add loc info & Fix api changes for CuTeDSL 4.4
#2204 opened Jan 23, 2026 by keithzzzzz Loading…
2
[Cute, SM100] Fix comment in tmem_p_offset
#2201 opened Jan 22, 2026 by Edenzzzz Loading…
Warn when ninja is missing
#2191 opened Jan 17, 2026 by blueberrycongee Loading…
1
Fix compute_block_sparsity import in benchmark_mask_mod
#2190 opened Jan 17, 2026 by blueberrycongee Loading…
1
[Cute][Testing] Protyping a fast test mode for Cute
#2188 opened Jan 16, 2026 by drisspg Loading…
1
[Cute,Fwd,Sm100] support irregular qhead / kvhead ratios
#2186 opened Jan 16, 2026 by timmy-feng • Draft
[Cute] Add torch.compile support for FA4
#2164 opened Jan 9, 2026 by gilfordting Loading…
8
Update mha_fwd.cpp, Normalize the commented-out parameters
#2160 opened Jan 9, 2026 by breakfei Loading…
ProTip! Find all pull requests that aren't related to any open issues with -linked:issue.