Deletion Tuning：只清零不到0.05%的权重即可匹敌全量微调

在遗忘问题上差异更明显：强化学习设置下，Deletion Tuning 的先前任务准确率保持在接近基座模型水平，而 FFT、LoRA 和随机稀疏微调最多损失 5.4 个点。为解释仅靠掩码为何足够，作者用 stable rank、谱漂移和主权重重叠三个几何指标分析强化学习下的权重变化，发现 Deletion Tuning 在三项指标上都最接近 FFT。

这项结果对持续学习有直接意义：顺序训练多个能力时可以在极小存储开销下进行，且比 LoRA 等主流参数高效方案更少遗忘，适合需要多任务后训练的工程团队参考。

The Surprising Effectiveness of Deleting Weights in LLM Post-Training

How little of a pretrained large language model (LLM) has to change for it to acquire new capabilities? We find that zeroing fewer than 0.05% of its weights, with no other modification, matches full fine-tuning (FFT) at scale across three LLM post-training regimes: reinforcement learning with verifiable rewards, supervised fine-tuning, and on-policy distillation. Our method, Deletion Tuning, learns a binary mask over the gate projection of transformer blocks. Two practical advantages follow: the trained mask is hundreds to thousands of times smaller than the corresponding FFT checkpoint, and Deletion Tuning forgets far less, which addresses a central challenge in continual learning. Under reinforcement learning, its prior-task accuracy stays near the base level while FFT, LoRA, and random sparse fine-tuning lose up to 5.4 points. To understand why masking alone suffices, we analyze the geometry of the learned weight changes under reinforcement learning using stable rank, spectrum drift, and principal-weight overlap. Compared with LoRA and random sparse fine-tuning, Deletion Tuning is closest to FFT across the three measures.

本条情报归属 AITOP「论文」核心栏目，重点聚焦北美前沿实验室与硅谷科技巨头。在跨区域产业观察坐标中，该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线，反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括：Deletion Tuning、LoRA、微调、强化学习，读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。

信源核验说明：本文原始线索采编自「arXiv cs.LG」（@Jack Lu, Zhenbang Yang, Mike Lasby, Tejas Pote, Yani Ioannou, Mengye Ren），经 AITOP 多维评分管线结构化提炼。