赞
踩
forward效果
AlexNet模型,
不Quantization | Q tilt (ReLU) | sign tilt (hard tanh) | Q(HWGQ) + ReLu backward | sign + ReLU backward | |||
Alex Net | Full Weight | top 1 | 55.7 | 55.7 | 46.7 | 49.5 | |
top 5 | 79.3 | 79.3 | 71< |
Copyright © 2003-2013 www.wpsshop.cn 版权所有,并保留所有权利。