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

ICLROral2026

Authors
Changxin Tian, jiapeng wang, Qian Zhao, Kunlong Chen, Jia Liu, Ziqi Liu, Jiaxin Mao, Xin Zhao, Zhiqiang Zhang, JUN ZHOU
Affiliation
Ant Group
Venue
ICLR 2026
Track
Oral

TL;DR

Recent advances in learning rate~(LR) scheduling have demonstrated the effectiveness of decay-free approaches that eliminate the traditional decay phase while maintaining competitive performance. Model merging techniques have emerged as particularly promising solutions in this domain.

Opening excerpt from the authors’ abstract. source

Read the paper

Topics

pre-training llm

← All ICLR 2026 Oral papers · Browse the whole archive