Skip to content

Pull requests: Dao-AILab/flash-attention

New pull request New
199 Open 671 Closed
199 Open 671 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

Fa4 return lse consistency
#2700 opened Jul 9, 2026 by yijianggit Loading…
[CuTe,Bwd,Sm100] Support head_dim not a multiple of 32 in backward
#2698 opened Jul 9, 2026 by KaijingOfficial Loading…
3 of 4 tasks
1
add tcgen.ld.red support to sm103a arch
#2696 opened Jul 6, 2026 by drisspg Collaborator Loading…
4
Validate VisionTransformer norm_layer
#2695 opened Jul 5, 2026 by dannyward630 Loading…
1
1
[CuTe] Fix fp8 P saturation in sm100 forward: reduce softmax max_offset 8 -> 4
#2694 opened Jul 3, 2026 by lishunyang12 Loading…
1
add linearize scheduling to combine kernel for full cudagraph
#2692 opened Jul 1, 2026 by liangel-02 Contributor • Draft
1
[Cute,Sm100] Drop redundant elect_one around bulk cute.copy (single-issue atoms)
#2689 opened Jun 29, 2026 by dongxiao92 Loading…
2
Make dropout-disabled FA2 libtorch ABI stable
#2688 opened Jun 28, 2026 by janeyx99 Contributor Loading…
4
[ROCm] Expose AITER_USE_SYSTEM_TRITON to preserve system Triton for flash-attn builds
#2687 opened Jun 27, 2026 by sstamenk Contributor Loading…
1
Expand FLASHATTENTION_DISABLE_DROPOUT to not bring in unneeded headers
#2669 opened Jun 20, 2026 by janeyx99 Contributor Loading…
5
Fix version comparison, contiguity checks, and IndexPutFirstAxis perf
#2663 opened Jun 17, 2026 by Kevin-Li-2025 Contributor Loading…
Enable 2CTA for SM100 block-sparse backward
#2661 opened Jun 16, 2026 by drisspg Collaborator Loading…
12
[NVIDIA][CuTe,Fwd/Bwd,sm120] Attention dropout (#2436), forward + backward
#2657 opened Jun 15, 2026 by johnnynunez Contributor Loading…
4
[CuTe,Bwd,Flex,sm80/sm120] Add full score_mod(_bwd) support
#2654 opened Jun 13, 2026 by larsmic Loading…
[CuTe,Bwd] Include subtile_factor in backward compile keys
#2645 opened Jun 11, 2026 by leonardHONG Loading…
[CuTe,Bwd,Sm90] fix: add missing griddepcontrol_wait in block-sparse producer
#2643 opened Jun 10, 2026 by codecaution Loading…
1
Harden NVIDIA toolchain extraction in hopper/setup.py against local attack
#2641 opened Jun 10, 2026 by AAtomical Loading…
FA4 consumer Blackwell (sm_120) integration
#2634 opened Jun 8, 2026 by thad0ctor Loading…
6
[CuTe,Bwd,Sm100] support non-native head dims via store predication
#2633 opened Jun 8, 2026 by WonsukYangTrlbs Loading…
chore: improve flash-attention maintenance path
#2630 opened Jun 5, 2026 by lphuc2250gma Loading…
Improve the benchmark script.
#2628 opened Jun 5, 2026 by xsank Loading…
[FA4] Add fused split_size prefill path for SM100
#2626 opened Jun 4, 2026 by Luosuu Contributor • Draft
Fix seqlen_k_loaded treating a window bound of 0 as unbounded (sliding window)
#2624 opened Jun 4, 2026 by lollinng Loading…
1
Validate paged-KV host config (tile_n divisible by KV-load thread count)
#2615 opened Jun 1, 2026 by fxdv Loading…
1
Enable pack_gqa backward on SM100/SM110
#2610 opened May 31, 2026 by fxdv Loading…
3
ProTip! Updated in the last three days: updated:>2026-07-06.