Search results

[FA4-FP8] on the B300. Low utilization of TC FMA XFU during inference.

Status: Open.

#2701 In Dao-AILab/flash-attention;

· Romaosir opened on Jul 10, 2026

[FA4] FA4 performs about the same as FA2 on B300.

Status: Open.

#2699 In Dao-AILab/flash-attention;

· hewangsh opened on Jul 9, 2026

Make dropout-disabled FA2 ABI stable with libtorch + cpython

Status: Open.

#2691 In Dao-AILab/flash-attention;

· janeyx99 opened on Jul 1, 2026

[Cute,Bwd,Sm100] Sparse MLA backward: two findings about dPsum stride divisibility and redundant elect

Status: Open.

#2677 In Dao-AILab/flash-attention;

· dongxiao92 opened on Jun 26, 2026

_flash_attn_fwd/_flash_attn_bwd: DSLRuntimeError — window_size_left/right expects Int32 but gets plain Python int

Status: Open.

#2653 In Dao-AILab/flash-attention;

· TyGu1 opened on Jun 13, 2026

atomic_add_fp32: TypeError on nvvm.atomicrmw(res=T.f32(), ...) — 'res' kwarg removed in cutlass-dsl 4.5.2

Status: Open.

#2652 In Dao-AILab/flash-attention;

· TyGu1 opened on Jun 13, 2026

FlashAttentionBackwardSm80: compute_softmax_scale_log2 overwrites softmax_scale with None → DSLRuntimeError

Status: Open.

#2651 In Dao-AILab/flash-attention;

· TyGu1 opened on Jun 13, 2026

_flash_attn_bwd: UnboundLocalError for dQ_single_wg on SM_12x (arch//10 == 12)

Status: Open.

#2650 In Dao-AILab/flash-attention;

· TyGu1 opened on Jun 13, 2026

FlashAttentionForwardSm120: use_tma_O incorrectly True on SM_121 → AttributeError on tma_atom_O=None

Status: Open.

#2649 In Dao-AILab/flash-attention;

· TyGu1 opened on Jun 13, 2026

Can flash-attn 2.8.3 be built from source with PyTorch 2.11.0 + CUDA 13.0 on A800 (sm80)?

Status: Open.

#2638 In Dao-AILab/flash-attention;

· SpareNoEfforttt opened on Jun 10, 2026

Bare tarfile.extractall into Predictable Cache During Build in flash-attention

Status: Open.

#2637 In Dao-AILab/flash-attention;

· AAtomical opened on Jun 10, 2026

Performance Optimization: Fusing Pre-Attention Normalization to bridge the unfused GatedMLP / Layer bounds

Status: Open.

#2632 In Dao-AILab/flash-attention;

· Tobi-Adesoye opened on Jun 8, 2026