nki(sigmoid): NKI (Trainium) implementation - #301

Open

bowencui123 wants to merge 10 commits into

mainfrom

bowen/nki/sigmoid

Commits on Aug 29, 2026

nki(sigmoid): NKI (Trainium) implementation

committed
3648403
nki(sigmoid): NkiAutotuner wiring (`block_size`)

bowencui123
and
claude
committed
67491ce

Commits on Aug 31, 2026

Merge branch 'main' into bowen/nki/sigmoid
Cecilia123li
authored
28b74bb
nki autotune results for sigmoid
Ubuntu
committed
6081635

Commits on Sep 9, 2026

nki(sigmoid): add NKI benchmark results (trn2.3xlarge, LNC2)

Ubuntu
and
claude
committed
ae1deb5

Commits on Sep 15, 2026

Merge remote-tracking branch 'origin/main' into bowen/nki/sigmoid
bowencui123
committed
a4df23f

Commits on Sep 16, 2026

nki(sigmoid): LNC2 flat kernel with in-kernel tails; XLA torch baseline on a padded 2-D view (1-D non-128-multiple activations are chunked by the compiler); rerun benchmarks

bowencui123
and
claude
committed
43222a1

Commits on Sep 21, 2026

Merge remote-tracking branch 'origin/main' into bowen/nki/sigmoid
bowencui123
committed
d076067
nki(sigmoid): migrate to the tilebench package layout

bowencui123
and
claude
committed
5ebfb18

Commits on Sep 22, 2026

nki(sigmoid): drop docstrings and inline comments

bowencui123
and
claude
committed
91fb219