English(EN) MuonGS: Pretraining LLMs with Relaxed Weight-Norm Control

MuonGS 通过新颖的权重范数控制增强大语言模型预训练

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-28 04:00

研究人员推出了一种新颖的大语言模型预训练方法 MuonGS，该方法通过宽松的权重范数控制来增强优化器性能。该方法使用累积有效学习率分析范数控制，并证明 MuonGS 通过使用 Gram--Schmidt 投影，保证了持续的范数累积和更快的收敛速度。在参数量从 1.35 亿到 10 亿的模型上进行的实验表明，与 MuonH 等现有方法相比，MuonGS 实现了更低的验证困惑度，并估计其加速比高达 15%。 AI

影响引入了一种新颖的优化技术，可能带来更高效的大语言模型训练和性能提升。

排序理由详细介绍大语言模型预训练新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 14 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 详细介绍大语言模型预训练新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, infra; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.AI TIER_1 English(EN) · Athanasios Glentis, Dawei Li, Mingyi Hong · 2026-10-28 04:00

MuonGS：具有宽松权重范数控制的预训练LLM

arXiv:2610.37647v1 Announce Type: cross Abstract: Weight decay and explicit weight-norm constraints improve optimizer performance in language model pretraining, yet a unified understanding of why such control helps and how much is needed remains incomplete. We analyze norm contro…

本文如何被排名

报道来源 [1]

MuonGS：具有宽松权重范数控制的预训练LLM

相关实体

相关话题