Issues 共 1262
_flash_attn_fwd/_flash_attn_bwd: DSLRuntimeError — window_size_left/right expects Int32 but gets plain Python int
#2653 · TyGu1 · 2026-06-13
Training Extremely Slow on Qwen3.5-35B-A3B + 8×B300 (280s/step), py-spy Shows FlashAttention-4 CUTLASS JIT Compilation During Backward
#2635 · yszhli · 2026-06-09
[FA4] cute varlen backward: int32 overflow in `_bwd_preprocess` corrupts `dQ/dK/dV` when `dout` has a large outer stride
#2611 · zhipenggong · 2026-06-01
buy crypto with credit card without OTP verification
#2607 · trufflez565 · 2026-05-31
[FA4, CuTe, SM100] `_flash_attn_fwd` fails at launch for `head_dim < 32` on B200
#2591 · rm-wu · 2026-05-25
[ROCm] gfx1201 compilation failure - v_cmp_u_f32 invalid
#2588 · UFO0506 · 2026-05-25
[FA4] Compilation too long for backward on TransformerEngine 2.15
#2571 · robinren03 · 2026-05-17
FA4 head_dim=256 B200 Regression Check (2026-05-16) with 2026-04-23
#2569 · Johnsonms · 2026-05-16
[Bug] [FA4] [hdim 256] CUDA error 700 in backward pass when using zero-length q sequences with hdim=256 in varlen mode
#2562 · umiswing · 2026-05-13
[FA4] [head_dim=256] dV is incorrect when using non-contiguous q, k, v from a single qkv projection
#2541 · vladrfl · 2026-05-05