Skip to content

Pull requests: Dao-AILab/flash-attention

New pull request New
147 Open 543 Closed
147 Open 543 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

layers/rotary: add rope_scaling support (linear, NTK, YaRN)
#2451 opened Apr 11, 2026 by GitGlimpse895 Loading…
Support batch invariant forward for FA3
#2450 opened Apr 10, 2026 by Edenzzzz Loading…
1
[Perf] SM103 tcgen05.ld.red for fused TMEM load + row-max in softmax
#2449 opened Apr 9, 2026 by LopezCastroRoberto Loading…
2
expose num_splits for FA2 and add option for kernel blocksize alignment
#2448 opened Apr 8, 2026 by liangel-02 Contributor Loading…
1
Add forward and backward support for FA4 hdim=256 on SM100
#2447 opened Apr 8, 2026 by wenkechen Loading…
[Cute,Fwd,Sm90] Ceil div in paged kv manager to prevent size 0
#2446 opened Apr 8, 2026 by imbr92 Contributor Loading…
[FA3][Hopper] Make the split heuristic sequence-aware for low-tile decode
#2445 opened Apr 8, 2026 by mllopartbsc Loading…
[Cute,Sm100,Fwd] add MLA 64/512 with topk sparsity for MQA 128 heads
#2441 opened Apr 6, 2026 by jayhshah Collaborator Loading…
Add dropout support to CuTe DSL attention kernels
#2439 opened Apr 6, 2026 by blake-snc Contributor Loading…
10 of 12 tasks
10
Draft: native FA4 bring-up for SM100 larger attention head dims
#2435 opened Apr 3, 2026 by m0at • Draft
[CuTe,Fwd,SM90] Enable head dim 512 for SM90
#2422 opened Apr 1, 2026 by IwakuraRein Loading…
6
[CuTe, SM120] Fix forward/backward bugs and add SM120 test guards
#2420 opened Apr 1, 2026 by 2imi9 Loading…
3
Add compress_factor for compressed causal attention
#2418 opened Mar 31, 2026 by jduprat Contributor Loading…
4
Fix SM120 forward pass crash: parent __init__ overwrites arch, enabling unsupported TMA path
#2416 opened Mar 31, 2026 by moghon92 Loading…
[Cute,Fwd,Sm90] Support SplitKV
#2415 opened Mar 31, 2026 by imbr92 Contributor Loading…
1
Feat([FA4][CUTE DSL]) Add head_dim=256 support (forward + backward)
#2412 opened Mar 30, 2026 by wangsiyu Loading…
23
chore(tests): move benchmarks to benchmarks/cute/ and reduce test prints
#2408 opened Mar 29, 2026 by NJX-njx Contributor Loading…
3 tasks
1
1
fix(flash_fwd_sm90): zero partial V smem to prevent 0*NaN=NaN in PV GEMM
#2407 opened Mar 29, 2026 by NJX-njx Contributor Loading…
3 tasks
1
[CuTe, SM120] Forward kernel with optimized TMA path and full features support
#2406 opened Mar 28, 2026 by sisgrad Loading…
6
feat: setup_context for FlashAttnFunc (torch.func.grad)
#2405 opened Mar 28, 2026 by NJX-njx Contributor Loading…
1
4
fix(cute): SM120 forward/bwd and atomic add compatibility
#2404 opened Mar 28, 2026 by NJX-njx Contributor Loading…
1
1
build(windows): MSVC conforming preprocessor for CUDA 13+ and ninja warning
#2403 opened Mar 28, 2026 by NJX-njx Contributor Loading…
1
1
[Cute,Sm80,Fwd] Guard kernel body with tile validity check to fix varlen illegal memory access
#2391 opened Mar 25, 2026 by zhuochenKIDD Contributor Loading…
1
Add SM80/SM120 block-sparse forward attention support
#2389 opened Mar 25, 2026 by blake-snc Contributor Loading…
6
Fix missing seqlen_info param in softcap scoremod
#2366 opened Mar 17, 2026 by rucnyz Loading…
ProTip! Mix and match filters to narrow down what you’re looking for.