Skip a Layer or Loop It? Learning Program-of-Layers in LLMs
跳过一层还是循环它?学习LLM中的层程序
机构 * Ziyue Li, Yang Li, Tianyi Zhou(李子悦、李阳、周天毅)
专题命中 推理与问题求解 :LLM(title_cn,abstract);large language model(abstract);language model(abstract);分类 cs.LG
AI总结 本文提出PoLar方法,通过轻量级预测网络学习为每个输入动态跳过或重复预训练层,在数学推理任务上以更少层数提升准确率。
Comments Accepted at ICML 2026. Substantially extends arXiv:2507.07996. Updated DART-Math experiments by removing duplicate examples; conclusions remain unchanged. Code: https://github.com/tianyi-lab/PoLar