发表机构
University of Michigan; University of Chicago（密歇根大学; 芝加哥大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对个性化场景下LLM微调成本过高的问题，提出无需权重更新的WFT方法，在LaMP基准上性能优于多数基线，仅用不到7%计算量接近SFT效果。
AI 中文摘要
监督微调（SFT）是使大语言模型（LLM）适配目标分布的标准方法，但在个性化等场景中，每个作者都需要单独的权重访问、优化、存储和重新训练，其成本变得高得难以承受。我们提出了无权重微调（WFT），这是一种无需训练的解码时方法，可在不更新权重的情况下近似SFT的分布效应。WFT会计算某一作者训练序列上的监督残差，并通过由dropout诱导的交叉协方差估计得到的跨前缀迁移算子，将这些残差迁移至当前提示词；该算子能捕捉某一上下文的扰动如何传播至另一上下文的预测，从而用对数几率空间修正替代基于梯度的参数更新。在三个LaMP个性化基准上，WFT取得了数据集间的最佳平均性能，在单个任务上与SFT相当或优于SFT，且在平均水平上优于其他轻量级基线。在预算控制的对比中，WFT仅用不到7%的有效计算量就达到了接近SFT的性能。对数几率层面的分析显示，在超过95%的下一个词概率质量上，WFT和SFT诱导的对数几率偏移的余弦相似度为0.875，这表明WFT无需修改模型权重即可捕捉监督适配的分布效应。
英文摘要
Supervised fine-tuning (SFT) is a standard approach for adapting LLMs to a target distribution, but in settings such as personalization, where each author requires separate weight access, optimization, storage, and retraining, its costs become prohibitive. We propose Weightless Fine-Tuning (WFT), a training-free decoding-time method that approximates the distributional effect of SFT without weight updates. WFT computes supervised residuals on an author's training sequence and transports them to the current prompt through a cross-prefix transport operator estimated from dropout-induced cross-covariance. The operator captures how a perturbation at one context propagates to predictions at another, replacing gradient-based parameter updates with logit-space corrections. On three LaMP personalization benchmarks, WFT achieves the best average performance across datasets, matches or exceeds SFT on individual tasks, and outperforms other lightweight baselines on average. In a budget-controlled comparison, WFT approaches SFT performance using less than 7% of the effective computation. Logit-level analysis shows a cosine similarity of 0.875 between the logit shifts induced by WFT and SFT over 95% of the next-token probability mass, suggesting that WFT captures the distributional effect of supervised adaptation without modifying model weights.
