Skip to content

Pull requests: vllm-project/flash-attention

New pull request New
Clear current search query, filters, and sorts
11 Open 68 Closed
11 Open 68 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

Pass s_aux through flash_attn_with_kvcache
#79 by tdoublep was merged Aug 8, 2025 Loading…
3
Attention Sinks Perf Boost
#78 by LucasWilkinson was merged Aug 9, 2025 Loading…
2
[WIP] Attention Sinks Perf Boost
#77 by LucasWilkinson was closed Aug 7, 2025 • Draft
1
[BugFix] Fix FA2 RuntimeError when sinks is provided
#76 by LucasWilkinson was merged Aug 6, 2025 Loading…
1
Support FA3 Attention Sink
#75 by zyongye was merged Aug 5, 2025 Loading…
cmake: get rid of empty VLLM_FA_GPU_ARCHES variable
#74 by dtrifiro was merged Jul 31, 2025 Loading…
1
Sparse attention : Generalize arch checks for A100 and above
#73 by ExtReMLapin was merged Jul 28, 2025 Loading…
5
[Misc] Add num_splits input arg to flash_attn_varlen_func
#72 by WoosukKwon was merged Jul 1, 2025 Loading…
1
vllm_flash_attn: Setup for vllm_kernels package
#71 by seemethere was merged Jun 23, 2025 Loading…
4
varlen combine scheduler
#70 by LucasWilkinson was merged Jun 16, 2025 Loading…
4
FA2 8.0 PTX
#69 by LucasWilkinson was merged Jun 16, 2025 Loading…
how are you supposed to run tests?
#68 by foolusion was closed May 22, 2025 Loading…
[BugFix] Fix raising exception when FA3 isn't available
#66 by LucasWilkinson was merged Apr 24, 2025 Loading…
1
Add rotary triton operator to vllm_flash_attn
#64 by cynthieye was merged Apr 24, 2025 Loading…
6
FA3 Decode Perf - Use single mma warp group for decode batches
#63 by LucasWilkinson was merged Apr 21, 2025 Loading…
2
[WIP] Use single mma warp group for decode batches
#62 by LucasWilkinson was closed Apr 18, 2025 • Draft
Upstream Sync - up to: d836a6bf09bf3838c6e71c9cf675b3708fea0d71
#61 by LucasWilkinson was merged Apr 10, 2025 Loading…
Sparse attention window size bug fix
#60 by mklasby was merged Apr 12, 2025 Loading…
3
[Easy] replace c10::optional with std::optional
#58 by yeqcharlotte was merged Mar 27, 2025 Loading…
3
Fix missing import in __init__
#57 by LucasWilkinson was merged Mar 25, 2025 Loading…
Upstream Sync - up to: 27f501dbe011f4371bff938fe7e09311ab3002fa
#56 by LucasWilkinson was merged Mar 20, 2025 Loading…
Avoid selecting fav3 for Blackwell
#55 by kushanam was merged Mar 5, 2025 Loading…
1
Fix building on CUDA 12.1
#53 by LucasWilkinson was merged Feb 27, 2025 Loading…
Upstream Sync | up to 06e34f62d18d3a721bc515d4b331a46d5d4c8c09
#52 by LucasWilkinson was merged Feb 26, 2025 Loading…
adding preliminary Blackwell support
#51 by kushanam was merged Mar 4, 2025 Loading…
4
ProTip! no:milestone will show everything without a milestone.