arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2603.04426cs.LGcs.AI
收藏

Delta-Crosscoder: 在狭窄微调条件下实现鲁棒的交叉编码器模型差异分析

Delta-Crosscoder: Robust Crosscoder Model Diffing in Narrow Fine-Tuning Regimes

Mila, Quebec AI Institute, Quebec, Canada(魁北克AI研究所);McGill University, Quebec, Canada(麦吉尔大学)

Aly Kassem, Thomas Jiralerspong, Negar Rostamzadeh, Golnoosh Farnadi

AI总结:

Delta-Crosscoder通过结合稀疏性和基于delta的损失,有效解决狭窄微调条件下的模型差异分析问题,优于现有方法。

摘要:

模型差异分析方法旨在识别微调如何改变模型的内部表示。交叉编码器通过学习基础模型和微调模型之间可解释的潜在方向共享字典来实现这一目标。然而,现有方法在狭窄微调条件下表现不佳,因为行为变化是局部且不对称的。我们引入了Delta-Crosscoder,它结合了BatchTopK稀疏性和基于delta的损失,优先考虑模型间变化的方向,以及通过匹配输入的激活对隐式对比信号。在10种模型生物体上进行评估,包括合成假事实、涌现的对齐偏差、潜意识学习和禁忌词猜测(Gemma、LLaMA、Qwen;1B-9B参数),Delta-Crosscoder能够可靠地隔离导致微调行为的潜在方向,并实现有效的缓解,优于基于SAE的基线方法,同时与非SAE方法相当。我们的结果表明,交叉编码器仍然是模型差异分析的强大工具。

英文摘要:

Model diffing methods aim to identify how fine-tuning changes a model's internal representations. Crosscoders approach this by learning shared dictionaries of interpretable latent directions between base and fine-tuned models. However, existing formulations struggle with narrow fine-tuning, where behavioral changes are localized and asymmetric. We introduce Delta-Crosscoder, which combines BatchTopK sparsity with a delta-based loss prioritizing directions that change between models, plus an implicit contrastive signal from paired activations on matched inputs. Evaluated across 10 model organisms, including synthetic false facts, emergent misalignment, subliminal learning, and taboo word guessing (Gemma, LLaMA, Qwen; 1B-9B parameters), Delta-Crosscoder reliably isolates latent directions causally responsible for fine-tuned behaviors and enables effective mitigation, outperforming SAE-based baselines, while matching the Non-SAE-based. Our results demonstrate that crosscoders remain a powerful tool for model diffing.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑