AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

WSM: Decay-Free Learning Rate Schedule via Checkpoint Merging for LLM Pre-training

Changxin Tian, jiapeng wang, Qian Zhao, Kunlong Chen, Jia Liu, Ziqi Liu, Jiaxin Mao, Xin Zhao, Zhiqiang Zhang, JUN ZHOU

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems · 19 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace learning-rate decay with checkpoint merging (WSM mean merge)WSD-annealed Ling-mini · Base model overall averageaverage score: 62.67 -> 63.95Improved
replace learning-rate decay with checkpoint merging, then SFTWSD-annealed Ling-mini · Instruct overall averageaverage score: 62.90 -> 64.07Improved
replace learning-rate decay with checkpoint merging on codeSFT model · Instruct Code averageaverage score: 58.23 -> 57.95Hurt
replace learning-rate decay with checkpoint merging on professional knowledgeBase model · Professional Knowledge averageaverage score: 53.46 -> 56.04Improved
use mean merge instead of EMA (ablation)WSM · Base overall averageaverage score: 63.01 -> 63.95Improved
use 1-sqrt merge weights instead of EMA (ablation)WSM · Base overall averageaverage score: 63.01 -> 64.06Improved
use 1-sqrt merge weights instead of mean (ablation)WSM · Base overall averageaverage score: 63.95 -> 64.06Improved
merge checkpoints instead of 1-sqrt LR decay (ablation)WSD · Base overall averageaverage score: 62.67 -> 64.06Improved
finer granularity (5B,16) instead of single checkpoint (80B,1) (ablation)WSM 80B merge · Base overall averageaverage score: 60.33 -> 63.63Improved
save every 10B and merge 8 instead of every 20B merge 4 (ablation)WSM 80B merge · Base overall averageaverage score: 63.36 -> 63.78Improved
larger merge window (ablation)WSM mean and 1-sqrt · Base benchmarksaverage score: gains diminish as window growsMixed
combine decay and merging (Decay-then-Merge or Merge-then-Decay) (ablation)WSM · Pre-training milestonesbenchmark scoreNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy