qazalandGitHub 62ad646d1c llama: gemm/fa backward speedups (gpt 5.6) (#17007) ...

* fp8 atb gemm speedup

* work

* revert

* fa bw faster

2026-07-14 00:27:00 +09:00

..

amd

llama: gemm/fa backward speedups (gpt 5.6) (#17007)

2026-07-14 00:27:00 +09:00

cuda

tk fa: use 16x64 tiles (#13086)

2025-11-03 18:25:38 -08:00

metal

feat: add thunderkittens (#12590)

2025-10-10 00:32:33 -07:00

tiny

full removal of dtype.vec (#16996)

2026-07-12 09:25:50 -07:00