CACTUS：曲率感知剪枝优化器，剪枝同时保留影响函数
Two For The Price Of One: Curvature-Aware Pruning Preserves Influence Functions Too
做剪枝的可以看看：CACTUS 比 LLM-Surgeon 快 28 倍，还能保留影响函数做数据选择。
论文提出 CACTUS，一个基于 ADMM 的剪枝优化器，目标函数同时考虑层激活和反向传播梯度，支持剪枝和稀疏加低秩两种压缩形式。在 Qwen3-4B 上做 50% 剪枝时，CACTUS 将与稠密模型的 C4 困惑度差距缩小 45%，WikiText-2 每样本梯度余弦相似度从 0.670 提升到 0.755。与针对同一目标的 LLM-Surgeon 相比，CACTUS 在更弱的硬件上运行快 28 倍以上。由于保留了影响函数，CACTUS 还能用于数据选择：在 Llama-3.2-3B 上，用它选出的 5% 数据微调新稠密模型，BIG-Bench Hard 准确率达 49.44%，对比 ALPS 的 47.22% 和稠密模型全量选择的 49.54%。
Two For The Price Of One: Curvature-Aware Pruning Preserves Influence Functions Too
Post-training pruning enables efficient deployment, but conventional methods do not explicitly preserve internal-gradient alignment between sparse and dense models. We study novel connections between downstream performance of compressed models and the ability to retain gradient information via influence-function preservation. Central to this connection is a curvature-aware objective that accounts for both layer activations and backpropagated gradients. Despite its promise, this objective remains underexplored in pruning due to a lack of efficient solvers. We introduce $\texttt{CACTUS}$, a novel ADMM-based optimizer supporting both pruning and sparse-plus-low-rank decompositions of pre-trained weights. Compared to LLM-Surgeon which targets the same objective, $\texttt{CACTUS}$ can achieve over an order-of-magnitude speedup. At $50\%$ pruning of Qwen3-4B, $\texttt{CACTUS}$ reduces the C4 perplexity gap to the dense model by $45\%$, improves compressed-to-dense per-sample gradient cosine similarity on WikiText-2 from $0.670$ to $0.755$, and runs over $28\times$ faster than LLM-Surgeon on substantially weaker hardware. By preserving influence functions, $\texttt{CACTUS}$ supports effective data selection. On Llama-3.2-3B, fine-tuning a fresh dense model on the $5\%$ of data selected by $\texttt{CACTUS}$ achieves $49.44\%$ accuracy on BIG-Bench Hard, compared with $47.22\%$ using ALPS and $49.54\%$ using dense-model selection. Finally, integrating $\texttt{CACTUS}$ with LoGRA, a recent gradient-projection strategy, on Qwen3-4B with MLP layers pruned to 2:4 sparsity and a rank-64 correction retains a $0.711$ cosine alignment with dense LoGRA-projected C4 gradients.
