A factored, low-memory optimizer for token embedding and LM-head tables matches Adam's quality while using kilobytes of optimizer state, and token trajectories appear nearly one-dimensional.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Token Geometry
A factored, low-memory optimizer for token embedding and LM-head tables matches Adam's quality while using kilobytes of optimizer state, and token trajectories appear nearly one-dimensional.