LoRA微调中，参数位置比数量更重要，选对位置才能有效提升性能。
Not How Many, But Which: Parameter Placement in Low-Rank Adaptation
- 根据梯度信息选择关键参数位置，而非随机分配。
- 在GRPO训练下，合理选位可达到标准LoRA精度，随机选位则无效。
- 只需不到0.5%成本快速识别核心参数，适用于1.5B到8B模型。
我们研究了低秩适配（LoRA）中的参数放置问题：在固定可训练参数数量为 $k$ 且矩阵 $A$ 冻结的前提下，这 $k$ 个参数的位置是否重要？在监督微调（SFT）下，随机与有指导的子集表现相当；但在基于基础模型的GRPO训练中，随机放置无法超越基线模型，而基于梯度信息的放置可恢复标准LoRA的准确率。这种差异源于梯度结构：SFT梯度为低秩且方向稳定，任意子集均能积累一致更新；而GRPO梯度为高秩且跨步骤近正交，仅持续同号梯度的参数保留学习信号。我们的评分方法可在不到10秒内完成，成本低于训练总成本的0.5%。被选中的参数集中于残差流写入投影（V, O, Down），且在不同模型家族和规模（1.5B - 8B）间保持稳定。
原文摘要 · Abstract (English)
We study the \textit{parameter placement problem}: given a fixed budget of $k$ trainable entries within the B matrix of a LoRA adapter (A frozen), does the choice of which $k$ matter? Under supervised fine-tuning, random and informed subsets achieve comparable performance. Under GRPO on base models, random placement fails to improve over the base model, while gradient-informed placement recovers standard LoRA accuracy. This regime dependence traces to gradient structure: SFT gradients are low-rank and directionally stable, so any subset accumulates coherent updates; GRPO gradients are high-rank and near-orthogonal across steps, so only elements with consistently signed gradients retain the learning signal. Our scoring procedure identifies these critical parameters in under 10 seconds at less than 0.5% of training cost. Selected parameters concentrate on residual-stream-writing projections (V, O, Down), stable across model families and scales (1.5B - 8B).
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。