SanSi: Looped Layers Boost Small Decision Models

The SanSi paper shows that reusing looped layers in typed decision models lets a 1.4B model gain 13.5 percentage points in accuracy, rivaling models three times its size.

2026-10-14 ~ 2026-10-14 · 2 related posts

1 near-duplicate retellings: rohanpaul_ai