Preprint
Jul 2026
Grokking Is Conditional and Fragile: A Fully-Tractable, Multi-Seed Study at 12K Parameters
A publicly released ~11,856-parameter Llama-style Llama-style transformer on modular arithmetic is studied, small enough to enumerate its weights, attention, and full input-output map, and grokking is measured as a multi-seed rate rather than a single outcome.
Yoshiyuki Ootani
· 0 citations