ViperEkura 9aca62c26c perf: remove per-element sentinel checks in softmax + fix stale comments ... - Replace 4*NC8 per-element -FLT_MAX comparisons with 2 row-level pn guards; masked entries naturally underflow via expf(-FLT_MAX - nm) ≈ 0; pn only guards all-masked-row edge where nm == -FLT_MAX (exp(0)=1 not 0); ~1-3% speedup on prefill (verified via standalone CUDA bench); correctness verified: prefill 4/4, decode 3/3, paged decode 13/13 - Fix stale comments: remove false 'pre-scale Q' claim, correct occupancy numbers, remove phantom sQ from smem description		2026-07-12 15:45:08 +08:00
..
kernels	perf: remove per-element sentinel checks in softmax + fix stale comments	2026-07-12 15:45:08 +08:00
tests	refactor: remove ineffective __launch_bounds__ from prefill kernel	2026-07-12 15:00:53 +08:00
__init__.py	feat: add optional CUDA kernel system (csrc/) + fused GQA decode attention	2026-07-06 12:09:58 +08:00
build.py	feat: paged decode attention with split-KV (scalar + MMA)	2026-07-11 18:14:51 +08:00