ViperEkura 21ddead238 fix: stabilize paged decode attention kernels ...

- zero-fill split partials so combine skips unwritten splits deterministically
- skip loading masked KV in paged decode kernels to avoid 0*NaN output poisoning
- zero-fill shared memory tile buffers to prevent stale NaN leaking into softmax

2026-07-31 21:01:12 +08:00

..

kernels

fix: stabilize paged decode attention kernels

2026-07-31 21:01:12 +08:00

tests

feat: unify attention backend with multi-dim mask support

2026-07-30 20:38:34 +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

refactor: unify rotary embedding interface and update docs

2026-07-31 16:52:25 +08:00