SpenseGPT:一次性剪枝实现LLM推理的稀疏与密集GEMM
一次剪枝,推理快两倍
SpenseGPT提出一种实用的一次性剪枝方法,在LLM推理中同时支持稀疏和密集通用矩阵乘法(GEMM)。该方法无需重新训练,仅通过一次剪枝即可大幅减少模型参数。
实验表明,在保持模型精度的前提下,剪枝后模型推理速度提升可达2倍。该技术适用于多种主流LLM架构。
SpenseGPT Practical One-shot Pruning Enabling Sparse and Dense GEMMs for LLM Inference📊 2 ⚡
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include SpenseGPT, 剪枝, LLM推理, GEMM.
Source attribution: curated from AK (@_akhaliq).