跳到正文
HuggingFace Daily Papers(社區熱門論文)·· 15 天前AI 評分46

刪除權重在 LLM 後訓練中的驚人效果

The Surprising Effectiveness of Deleting Weights in LLM Post-Training

AI 導讀

Deletion Tuning 在 LLM 後訓練中只將少於 0.05% 的權重歸零,便在可驗證獎勵強化學習、監督式微調及 on-policy 蒸餾三種設定下,達到與全量微調(FFT)相當的效果。其二元遮罩比相應 FFT checkpoint 小數百至數千倍,遺忘程度亦較低。強化學習中,其先前任務準確率維持接近基礎模型;FFT、LoRA 及隨機稀疏微調則最多下降 5.4 個百分點。

正文

Published on Nov 7

Authors:

,

,

,

,

Abstract

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.

View arXiv page View PDF Project page GitHub Add to collection

Models citing this paper 1

Jacklu0831/deletion-tuning-masks

Updated about 21 hours ago

•

6

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2611.10866 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2611.10866 in a Space README.md to link it from this page.

Collections including this paper 0

No Collection including this paper

Add this paper to a collection to link it from this page.

來源:HuggingFace Daily Papers(社區熱門論文) · huggingface.co