New LoRA Skills Should Read but Never Write
作者: Zeyan Li, Panqi Yang, Qirong Guo, Shengda Zhuo, SIyuan Qiu, Hu Xu, Chun Li, Jianfeng Xu
分类: cs.LG
发布日期: 2026-09-25
💡 一句话要点
提出READ方法以解决LoRA适配器组合问题
🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)
关键词: 低秩适配器 技能组合 模型微调 自然语言处理 多任务学习
📋 核心要点
- 现有的LoRA适配器组合方法在权重合并时容易产生干扰，且重新训练成本高。
- 本文提出READ方法，通过重写适配器为标准形式，确保更新的精确性，并限制耦合方向。
- 实验结果显示，READ在SuperGLUE和领域基准测试中相较于现有基线有显著提升，平均提高超过20分。
📝 摘要（中文）
低秩适配器（LoRA）使得针对每个任务微调大型语言模型变得经济，但将多个独立训练的适配器合并为一个模型仍然困难。现有方法在权重空间合并更新时会导致干扰，重新训练所有任务数据成本高昂，而在不同适配器之间进行路由则放弃了单一组合模型的目标。本文提出READ（只读适配器增量扩展），通过将每个适配器重写为平衡的标准形式，确保其更新被精确保留，并且耦合仅在一个方向上增长，从而使新技能能够读取旧技能的输入子空间但不能写入其输出子空间。实验表明，READ在多个基准测试中显著提升了性能。
🔬 方法详解
问题定义：本文解决的是如何有效地将多个独立训练的LoRA适配器合并为一个模型的问题。现有方法在合并权重时容易产生干扰，且重新训练所有任务数据的成本过高。
核心思路：论文提出的READ方法通过将每个适配器重写为平衡的标准形式，确保其更新被精确保留，并且耦合仅在一个方向上增长，从而避免了旧技能的输出被新技能覆盖。
技术框架：READ的整体架构包括适配器的重写模块和耦合矩阵的更新模块。每个适配器在重写后保持其更新的完整性，而新技能的耦合矩阵行是唯一可训练的部分。
关键创新：READ的主要创新在于其耦合方向的单向性和适配器更新的标准化，这与现有方法的多向耦合和不确定更新方式形成鲜明对比。
关键设计：在READ中，适配器的重写过程确保了更新的平衡性，耦合矩阵的设计使得新技能只能读取旧技能的输入而不能修改其输出，避免了信息干扰。
🖼️ 关键图片
📊 实验亮点
在实验中，READ方法在SuperGLUE基准测试中平均提高了超过20分，在领域基准测试中提高了超过7分，几乎所有完整的添加序列都超越了每个直接基线，展示了其在技能组合中的显著优势。
🎯 应用场景
该研究的潜在应用领域包括自然语言处理中的任务适配、对话系统的技能扩展以及多任务学习等。通过有效地组合不同的技能，READ方法可以提升模型在特定任务上的表现，具有重要的实际价值和广泛的应用前景。
📄 摘要（原文）
Low-rank adapters (LoRA) make it cheap to fine-tune a large language model once per task, but combining several independently trained adapters into one model remains difficult: merging the updates in weight space causes interference, retraining on all task data is expensive, and routing between separate adapters gives up the goal of a single combined model. We trace the difficulty to two choices that every composition method makes implicitly. A LoRA update admits infinitely many equivalent factorizations; the choice among them is invisible while an adapter serves alone, but it determines what a learned interaction between adapters can see. A coupling between an old skill and a new one can likewise point in either direction, and the direction decides whether the old skills keep computing what they computed before. We introduce READ (Read-only Expansion of Adapter Deltas), which fixes both choices: each adapter is rewritten into a balanced canonical form that preserves its update exactly, and the coupling grows in one direction only, so a new skill can read the input subspaces of old skills but cannot write into their output subspaces. The only trainable object at each append is the new skill's row of the coupling matrix, and the composed update folds into the base weights with no inference cost, routing, or task-specific rules. We evaluate READ across four benchmark suites and two model families, adding skills one at a time. Across several families, READ improves every suite average over the strongest published baselines built from the same adapters---by more than twenty points on SuperGLUE and more than seven points on the domain suite---and nearly all complete addition sequences end above every direct baseline. Factor coordinates and coupling direction, which a lone adapter never exposes, are what decide whether composed skills survive.