BriefGPT.xyz
May, 2021
BatchQuant: 鲁棒量化器的量子化全架构搜索
BatchQuant: Quantized-for-all Architecture Search with Robust Quantizer
HTML
PDF
Haoping Bai, Meng Cao, Ping Huang, Jiulong Shan
TL;DR
我们提出了BatchQuant,这是一种稳健的量化器公式,可在数量少得多的GPU小时内训练出一种超过10^{76}个量化子网的紧凑超网,并首次无需重新训练即可无缝扩展一次权重共享NAS超网以支持任意超低位宽混合精度量化策略的子网。
Abstract
As the applications of
deep learning
models on edge devices increase at an accelerating pace, fast adaptation to various scenarios with varying resource constraints has become a crucial aspect of model deployment. As a result,
→