Issues

• [Feature Request] [RFC] A shared, extensible interface for residual connections

  enhancement
  fla-org/flash-linear-attention#1332 · Filter by author feiyunzhang opened on Oct 6, 2026

• [Bug] chunk_bwd_kernel_dv_local is nondeterministic at num_warps=8 on Blackwell (same race as #1228)

  bug
  fla-org/flash-linear-attention#1327 · Filter by author kaloyan-inherent opened on Oct 4, 2026

• [Perf] Gluon attention backend with large head dimensions

  fla-org/flash-linear-attention#1322 · Filter by author yzhangcs opened on Oct 3, 2026
  2

• [Bug] MesaNet gradients become inaccurate when the loss is scaled down

  fla-org/flash-linear-attention#1283 · Filter by author Zhubaoheng opened on Sep 24, 2026

• [Bug] Cache continuation drops inputs_embeds in FLAGenerationMixin

  fla-org/flash-linear-attention#1281 · Filter by author aniruddhaadak80 opened on Sep 24, 2026

• [RFC] [Bench] Deduplicate graph benchmark runner against the eager benchmark infrastructure

  fla-org/flash-linear-attention#1275 · Filter by author FightingZhen opened on Sep 22, 2026

• [Bug] SimpleGatedLinearAttention masks v for padding but not the decay gate gk

  fla-org/flash-linear-attention#1272 · Filter by author AmirF194 opened on Sep 21, 2026

• [Bug] BF16 Transformer dense/packed test fails with split-KV

  fla-org/flash-linear-attention#1255 · Filter by author Lexwzp0 opened on Sep 16, 2026

• [Bug] NPU triton-ascend kernels silently mismatch on unaligned K/V/T (CANN leftover DMA aliases the next row)

  ascend-npubug
  fla-org/flash-linear-attention#1251 · Filter by author sunyi0505 opened on Sep 15, 2026
  6

• [Bug] KDA safe_gate fast path can make prefix outputs and gradients depend on future gates

  needs-verification
  fla-org/flash-linear-attention#1240 · Filter by author ZTyler0303 opened on Sep 10, 2026
  4

• [Proposal] SM80 (A100/A800) CUDA backend for chunk_gated_delta_rule — optional dependency, FlashKDA-style

  fla-org/flash-linear-attention#1234 · Filter by author Leslie360 opened on Sep 9, 2026

• [RFC] Add Graph Capture Support for FLA Prefill and Training Paths on Ascend NPUs

  ascend-npuenhancement
  fla-org/flash-linear-attention#1233 · Filter by author FightingZhen opened on Sep 8, 2026
  1

• [Bug] tl.dot(..., allow_tf32=False) returns NaN/Inf in recompute_w_u_fwd_kernel_npu with fully-finite bf16 inputs

  ascend-npubug
  fla-org/flash-linear-attention#1215 · Filter by author TimeVShow opened on Sep 3, 2026
  4

• [RFC] Graph capture support for prefill and training: GDN/KDA chunk and convolution paths

  enhancement
  fla-org/flash-linear-attention#1155 · Filter by author zhiyuan1i opened on Aug 20, 2026
  4

• [RFC] Roadmap: NPU support for flash-linear-attention

  ascend-npuenhancement
  fla-org/flash-linear-attention#942 · Filter by author zheliuyu opened on Jun 9, 2026
  11

• [Feature Request] Introducing an input buffer for final_state in fused_recurrent_gated_delta_rule_fwd

  enhancement
  fla-org/flash-linear-attention#872 · Filter by author ir2718 opened on Apr 30, 2026
  3

• [Feature Request] Implement tensor inverse

  enhancement
  fla-org/flash-linear-attention#680 · Filter by author sustcsonglin opened on Dec 10, 2025

• [RFC] Context Parallel Implementations

  enhancement
  fla-org/flash-linear-attention#674 · Filter by author garrett361 opened on Dec 1, 2025
  3

  

• [RFC] Merge high similarity kernels between KDA, GLA and DPLR

  enhancement
  fla-org/flash-linear-attention#659 · Filter by author sustcsonglin opened on Nov 23, 2025
  1

• [RFC] Fuse L2 norm into short conv?

  enhancement
  fla-org/flash-linear-attention#657 · Filter by author sustcsonglin opened on Nov 22, 2025
  2

• [Feature Request] Gated Delta Net (Norm) Unfused Gate

  enhancement
  fla-org/flash-linear-attention#596 · Filter by author CoffeeVampir3 opened on Sep 18, 2025
  2

  

• [RFC] Support chunk prefilling for all kernels

  enhancementurgent
  fla-org/flash-linear-attention#502 · Filter by author sustcsonglin opened on Jul 1, 2025
  1

• [RFC] Add flag to control recomputation in chunk kernels

  enhancement
  fla-org/flash-linear-attention#485 · Filter by author sustcsonglin opened on Jun 25, 2025
  

• [RFC] Add stick-breaking attention

  enhancement
  fla-org/flash-linear-attention#481 · Filter by author yzhangcs opened on Jun 23, 2025
  3

• [RFC] Optimizing Triton kernel launching

  enhancement
  fla-org/flash-linear-attention#471 · Filter by author sustcsonglin opened on Jun 19, 2025
  2