Batch-wise Adaptive Pruning: Periodic Neuron Activation-Aware Weight Pruning for Language Reasoning Model
批量自适应剪枝:面向语言推理模型的周期性神经元激活感知权重剪枝
机构 * The University of Tokyo(东京大学)
专题命中 复杂问题求解 :reasoning(title,abstract);chain-of-thought(abstract);分类 cs.CL
AI总结 本研究针对语言推理模型批量推理下自适应剪枝性能下降的问题,提出带周期性top-k选择和激活记忆的无训练剪枝方法,在DeepSeek-R1-Distill-Qwen-7B上实现39.7个百分点准确率提升与1.40倍推理加速。
Comments Accepted at COLM 2026. 28 pages, 12 figures, 18 tables. Code: this https URL (https://github.com/matsuolab/batch-wise-prune)