研究剪枝对大模型解释工具的影响，发现剪枝会破坏解释能力。
When Pruning Meets Interpretability: Preserving Sparse Autoencoder Robustness in LLMs
- 用扰动能量理论分析剪枝对稀疏自编码器的影响
- 中层剪枝最敏感，激活感知方法更抗剪枝
- 提出分层稀疏策略，提升压缩后模型性能
稀疏自编码器（SAEs）广泛用于解析大语言模型（LLMs）的内部表征，但其在后处理剪枝下的可靠性尚不明确。本文系统研究了剪枝对SAE行为的影响，理论上证明：对固定SAE而言，其影响由一种协方差加权范数——扰动能量决定。该视角揭示了权重幅度剪枝的关键局限：忽略激活几何结构，扭曲学习到的表示空间，导致SAE功能退化。相比之下，如Wanda和SparseGPT等激活感知方法能隐式控制扰动能量，因而显著更鲁棒。我们进一步发现所有剪枝方法均存在一致的结构性脆弱：中间层远比首尾层对剪枝更敏感。基于此，提出分层稀疏分配策略，在相同平均剪枝率下实现更低困惑度。四个模型架构的实验验证了理论发现。代码已公开于https://github.com/osu-srml/sae-robustness-under-pruning/tree/main。
原文摘要 · Abstract (English)
Sparse autoencoders (SAEs) are widely used to interpret the internal representations of large language models (LLMs), yet their reliability under post-hoc model compression remains poorly understood. We present a systematic study of how pruning affects SAE behavior and theoretically show that, for a fixed SAE, its impact is governed by perturbation energy, a covariance-weighted norm. This perspective exposes a key limitation of magnitude pruning: by ignoring activation geometry, it distorts the learned representation space and degrades SAE functionality. Activation-aware methods such as Wanda and SparseGPT, in contrast, implicitly control perturbation energy and are therefore substantially more robust at preserving SAE behavior. We further reveal a consistent structural vulnerability across all pruning methods: middle layers are significantly more sensitive to pruning than early or late layers. Guided by this insight, we propose a layer-wise sparsity allocation strategy, achieving lower perplexity under the same average pruning sparsity. Experiments across four model architectures validate our theoretical findings. Code is publicly available at https://github.com/osu-srml/sae-robustness-under-pruning/tree/main.
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。