---
id: 20260804-T0-02
title: "LARA：残差流轻量适配器，参数效率超越LoRA"
title_en: "LARA: Lightweight Residual Adapters Outperform LoRA in Parameter Efficiency"
url: https://ai.daily.yangsir.net/daily/20260804-T0-02
issue_date: 2026-08-04
publish_date: 2026-08-03T04:00:00.000Z
category: research
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2607.28669
---

# LARA：残差流轻量适配器，参数效率超越LoRA

arXiv 新论文提出 LARA（轻量加性残差适配）方法。不同于 LoRA 在权重矩阵中添加低秩更新，LARA 在冻结模型的残差流中读取高维隐藏状态并添加可学习的加性更新。该方法在保持模型核心能力的同时，实现了高效的适配与对齐，且所需参数量少于 LoRA。

## English Version

**LARA: Lightweight Residual Adapters Outperform LoRA in Parameter Efficiency**

A new arXiv paper presents LARA (Lightweight Additive Residual Adaptation). Unlike LoRA, which adds low-rank updates to weight matrices, LARA reads the high-dimensional hidden state in a frozen model's residual stream and adds learnable additive updates. This achieves efficient adaptation and alignment while preserving core model capabilities, with fewer parameters than LoRA.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2607.28669)

**详情页**：https://ai.daily.yangsir.net/daily/20260804-T0-02

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*