feat: Triton fused SoftSignGLU kernel for LYNXNet2[WIP]#312

Draft

KakaruHayate wants to merge 17 commits into

openvpi:mainfrom

KakaruHayate:feat/fused-softsign-glu

Commits on Jul 2, 2026

feat: Triton fused SoftSignGLU kernel for LYNXNet2
KakaruHayate
committed
perf: skip variance fusion, realistic warmup M, fewer autotune configs
KakaruHayate
committed
fix: remove useless warmup from __init__ (model on CPU, no compilation)
KakaruHayate
committed
fix: disable variance fusion by default (small backbones, no benefit)
KakaruHayate
committed
fix: remove useless warmup from __init__ (model on CPU, no compilation)
KakaruHayate
committed
fix: address PR review issues
KakaruHayate
committed
fix: elem kernel extra N arg; variance nested glu_type
KakaruHayate
committed

Commits on Jul 3, 2026

fix: backward kernel dtype not hardcoded fp16
KakaruHayate
committed

Commits on Jul 7, 2026

Commits on Jul 14, 2026

review: fix hard blocks — TF32 removal, glu_type revert, robustness
KakaruHayate
committed

Commits on Jul 21, 2026

fix: fail fast when Triton is unavailable
KakaruHayate
committed