Skip to content

Pull requests: Dao-AILab/flash-attention

New pull request New
91 Open 360 Closed
91 Open 360 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

[Cute] Fix: arg pass in cute flash-attn inferface
#2101 opened Dec 27, 2025 by SeanLi-OI Loading…
[WIP,Cute,Fwd] improved block sparsity
#2100 opened Dec 26, 2025 by reubenconducts • Draft
Add pack-gqa fwd support for sparse impl w/ broadcasted H dim
#2098 opened Dec 24, 2025 by drisspg Loading…
Fix softmax incorrect row_max issue
#2083 opened Dec 17, 2025 by imbr92 Loading…
Fix TypeError when ColumnParallelLinear is None
#2080 opened Dec 17, 2025 by ailuntz Loading…
1
Reduce Chance of Build OOM
#2079 opened Dec 17, 2025 by Qubitium Loading…
1
[Cute,Fwd,SM100] Add softmax precision control parameters: rescale_threshold and disable_e2e
#2067 opened Dec 15, 2025 by ssxuwinter Loading…
7
Add missing code highlighting to the README
#2061 opened Dec 10, 2025 by bryant1410 Loading…
Update README.md
#2058 opened Dec 10, 2025 by eduardoruiz1999 Loading…
[AMD ROCm] Enable CK backend for ROCm gfx12
#2054 opened Dec 8, 2025 by hyoon1 Loading…
1
Fix Windows Linking for FlashAttention 3 using Ninja response files
#2047 opened Dec 4, 2025 by windreamer Loading…
3
Ko3n1g/ci/torch2.9 for cuda129
#2044 opened Dec 3, 2025 by ko3n1g • Draft
2
Disable abi3 for free-threaded python
#2034 opened Nov 25, 2025 by kevmo314 Loading…
[Cute,Fwd/Bwd,Sm12x] [WIP/DRAFT/HELP] cute FA for sm12x
#2017 opened Nov 16, 2025 by johnnynunez • Draft
6
[Cute,Fwd,Sm100] Support q_stage=1 for inference
#1993 opened Nov 7, 2025 by timmy-feng Loading…
3
[Cute,Fwd,Sm90] Support KV cache
#1992 opened Nov 6, 2025 by imbr92 Loading…
2
Local window attention deterministic mode efficiency improvement
#1972 opened Oct 29, 2025 by GD06 • Draft
Make wheel name and version consistent
#1956 opened Oct 22, 2025 by bobingm Loading…
1
fix: nan when m_i_new=-inf in online softmax
#1948 opened Oct 20, 2025 by tongyx361 Loading…
1
support cpu run fa triton kernel
#1938 opened Oct 15, 2025 by hellozmz • Draft
windows error error C2039:
#1932 opened Oct 12, 2025 by Granddyser Loading…
feat: add to support float8 kvcache in fa4
#1914 opened Sep 28, 2025 by yicwang Loading…
3
fix forward and backward kernel
#1907 opened Sep 24, 2025 by rz2778 Loading…
Add flash_attn_varlen_qkvpacked_func to hopper (flash_attn_3)
#1902 opened Sep 22, 2025 by foreverYoungGitHub Loading…
Feature/varlen rotray
#1899 opened Sep 19, 2025 by mhoangvslev Loading…
ProTip! Exclude everything labeled bug with -label:bug.