Abstract
Model pruning technique is vital for accelerating large language models by reducing their size and computational requirements. However, the generalizability of existing pruning methods across diverse datasets and tasks remains unclear. Thus, we conduct extensive evaluations on 24 datasets and 4 tasks using popular pruning methods. Based on these evaluations, we find and then investigate that calibration set greatly affect the performance of pruning methods. In addition, we surprisingly find a significant performance drop of existing pruning methods in sentiment classification tasks. To understand the link between performance drop and pruned neurons, we propose Neuron Semantic Attribution, which learns to associate each neuron with specific semantics. This method first makes the unpruned neurons of LLMs explainable.
Abstract (translated)
模型剪枝技术对于通过减少大型语言模型的规模和计算需求来加速这些模型至关重要。然而,现有的剪枝方法在不同数据集和任务中的泛化能力尚不清楚。因此,我们使用流行的方法对24个数据集和4种任务进行了广泛的评估。基于这些评估结果,我们发现并调查了校准集(calibration set)极大地影响了剪枝方法的性能。此外,我们惊讶地发现了现有剪枝方法在情感分类任务中的显著性能下降。为了理解性能下降与被剪枝神经元之间的联系,我们提出了“神经语义归属”(Neuron Semantic Attribution),该方法旨在将每个神经元与特定语义关联起来。这种方法首先使大型语言模型中未被剪枝的神经元变得可解释。
URL
https://arxiv.org/abs/2503.01542