Signals 4 · free daily AI digest

Rounding in Preconditioner Space: Redesigning 4-bit AdamW Optimizer-State Quantization

Paper recorded by Signals 4 on 2026-10-08 in cs.LG. Abstract reproduced from arXiv; link to the original below.

Published 2026-10-08 on arXiv · recorded by Signals 4 on 2026-10-09

Category: cs.LG · 机器学习 · first seen 2026-10-09

Abstract

Quantizing AdamW's optimizer states reduces persistent storage, but quantization errors propagate through the moment recurrences and perturb subsequent adaptive updates. We redesign 4-bit optimizer-state quantization for AdamW from the perspective of \emph{rounding space}: the coordinate in which a quantizer chooses between adjacent reconstruction levels. For the second moment, a local analysis of

Read on arXiv →

#318 most recent of 749 cs.LG papers we have recorded · ↑ newer: One Block, Multiple Depths: Recurrent Vision Transformers with Depth-P · ↓ older: Density Ratio Estimation with Stein Displacement Fields
Cite this page: Rounding in Preconditioner Space: Redesigning 4-bit AdamW Optimizer-State Quantization: the #318 most recent of 749 cs.LG papers we have recorded (as of 2026-10-08). Source: Signals 4 (Signals API) — https://data.jiangzhang.ca/signals4/t/papers/rounding-in-preconditioner-space-redesigning-4-bit-adamw-optimizer-state-quantiz.html
Free to quote with attribution to “Signals 4 (Signals API)”. Machine-readable: papers.json
Related: More cs.LG papers · arXiv signals · All papers · Today in AI
Get 4 AI signals a day by email — free.
Subscribe free → See all plans →
Get 4 AI signals a day by email — free
All models · All repos · By company · Daily editions