Skip to content

Pull requests: fla-org/flash-linear-attention

New pull request New
40 Open 795 Closed
40 Open 795 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

[Fix] Eliminate per-call reference cycle in launch_grid_chunked
#1235 opened Sep 9, 2026 by Ginray Loading…
4 of 5 tasks
[Ops] Extend fixed-capacity CUDA Graph support for GDN and KDA enhancement New feature or request needs-verification Lacks real execution evidence (CI skipped / no before-after data) waiting-author Reviewer acted; ball is in the author's court
#1230 opened Sep 7, 2026 by 1418036869abc-cmd Loading…
4 of 5 tasks
2
[Refactor] Bind backend dispatch to local registries enhancement New feature or request
#1223 opened Sep 6, 2026 by yzhangcs Member • Draft
4 of 5 tasks
10
[Fix] Correct packed loss boundaries and cache defaults bug Something isn't working
#1217 opened Sep 3, 2026 by taking-lying-flat Contributor Loading…
4 of 5 tasks
4
[Ops] Add Triton-Ascend compatibility for GDN-2 ascend-npu Ascend NPU (triton_ascend) related
#1213 opened Sep 1, 2026 by hazelduan Loading…
3 of 5 tasks
6
Gate solve_tril dot precision on whether TF32 is faster, not whether it exists bug Something isn't working waiting-author Reviewer acted; ball is in the author's court
#1205 opened Aug 30, 2026 by arbi-dev Loading…
[Rotary] Separate varlen lengths from cache capacity bug Something isn't working
#1204 opened Aug 30, 2026 by taking-lying-flat Contributor Loading…
4 of 5 tasks
2
[GDN] Add configurable output gate activation enhancement New feature or request minor Low-value-density change (typo/docs/small validation), batch-process
#1182 opened Aug 27, 2026 by alifurkanstahl Loading…
4 of 5 tasks
[Docs] Add software health index badge minor Low-value-density change (typo/docs/small validation), batch-process
#1153 opened Aug 19, 2026 by Nayjest Loading…
5 tasks done
[Ops] Use block matrix inversion to speed up solve_tril for the Ascend NPU backend ascend-npu Ascend NPU (triton_ascend) related needs-verification Lacks real execution evidence (CI skipped / no before-after data)
#1145 opened Aug 17, 2026 by OsirisDuan Contributor Loading…
4 of 5 tasks
3
[Test] Add realistic benchmark input profiles enhancement New feature or request
#1144 opened Aug 17, 2026 by 0z5a Loading…
5 tasks done
[Perf] Accelerate KDA training with TileLang needs-verification Lacks real execution evidence (CI skipped / no before-after data) performance
#1128 opened Aug 14, 2026 by lbx154 Contributor Loading…
5 tasks done
1
[Cache] Memoize kernel config lookups to avoid per-launch file access needs-verification Lacks real execution evidence (CI skipped / no before-after data)
#1123 opened Aug 13, 2026 by EastZeus Contributor Loading…
5 tasks done
[KDA] Warn on kwargs silently dropped by chunk_kda and fused_recurrent_kda minor Low-value-density change (typo/docs/small validation), batch-process
#1122 opened Aug 13, 2026 by EastZeus Contributor Loading…
5 tasks done
1
[Perf] Parallelize AttnRes long-sequence reduction needs-verification Lacks real execution evidence (CI skipped / no before-after data) performance
#1114 opened Aug 8, 2026 by lbx154 Contributor Loading…
5 tasks done
9
[KDA] Add FlashKDA CUDA training backend for chunk_kda needs-verification Lacks real execution evidence (CI skipped / no before-after data)
#1112 opened Aug 8, 2026 by xy200303 Contributor Loading…
5 tasks done
1
[GDN] feat(triton_ascend): add fused_recurrent GDN fwd kernel and dispatch ascend-npu Ascend NPU (triton_ascend) related enhancement New feature or request
#1078 opened Jul 30, 2026 by OsirisDuan Contributor Loading…
[KDA] Add TLE inference backend for NVIDIA H800 with up to 1.41x speedup needs-verification Lacks real execution evidence (CI skipped / no before-after data) performance
#1075 opened Jul 30, 2026 by kuoihao Loading…
2
[Perf] TileLang backends for linear-attention chunk backward (simple_gla / gla / delta_rule / gdn) waiting-author Reviewer acted; ball is in the author's court
#1046 opened Jul 19, 2026 by markovchain-builder Contributor Loading…
6
[Feature] support reusable final_state buffer in fused recurrent GDN enhancement New feature or request
#1043 opened Jul 18, 2026 by Bsgg1 Loading…
1
[Fix] Reject non-divisible GQA head counts bug Something isn't working minor Low-value-density change (typo/docs/small validation), batch-process
#1032 opened Jul 16, 2026 by morluto Contributor Loading…
1
1
[GDN] Add batch-invariant mode: bitwise-reproducible prefill/decode across batch sizes enhancement New feature or request
#1003 opened Jul 4, 2026 by yichuan-w Loading…
[GDN2] Add fused BT=16 inference kernels for GDN2 prefill enhancement New feature or request
#990 opened Jun 29, 2026 by Ghd-2077 Loading…
[Attn] Remove erroneous @torch.compile on ParallelAttentionFunction class bug Something isn't working
#980 opened Jun 25, 2026 by arbi-dev • Draft
1
[Perf] Generalize fused q/k/v short convolution across layers performance
#977 opened Jun 23, 2026 by zhiyuan1i Collaborator Loading…
4
ProTip! no:milestone will show everything without a milestone.