English(EN) LoRA-TSD: Tangent-Space Spectral Descent for LoRA via Muon-Style Updates

新的LoRA-TSD优化器为LLM提供更快、更鲁棒的微调

作者 PulseAugur 编辑部 · [1 个来源] · 2026-09-03 04:00

研究人员推出了一种新颖的大型语言模型微调优化器LoRA-TSD。该方法将每次更新视为固定秩矩阵流形上的切向量，并在该切空间内执行谱范数最速下降步。LoRA-TSD提供了一种比先前基于流形的方法便宜高达2.8倍的回缩方法，并在自然平稳性度量下为LoRA训练提供了首个全局收敛保证。在Llama-3.2-1B和Qwen3-32B等模型上的六个基准测试实验表明，LoRA-TSD的性能优于现有的LoRA优化器。 AI

影响这项新的优化技术可能导致更高效、更有效的大型语言模型微调，从而降低计算成本并提高下游任务的性能。

排序理由该集群包含一篇详细介绍大型语言模型微调新方法的论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 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: 22 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇详细介绍大型语言模型微调新方法的论文。[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.LG TIER_1 English(EN) · Dmitrii Andriianov, Andrey Veprikov, Aleksandr Beznosikov · 2026-09-03 04:00

LoRA-TSD：LoRA的切空间谱下降，通过μ子式更新

arXiv:2609.02734v1 Announce Type: new Abstract: Low-rank adaptation (LoRA) is the standard way to fine-tune large models, yet when its two factors are trained independently, the update ignores the geometry of the low-rank weight change it induces. We introduce LoRA-TSD, an optimi…

本文如何被排名

报道来源 [1]

LoRA-TSD：LoRA的切空间谱下降，通过μ子式更新

相关实体

相关话题