Record the forward and backward time

最新推荐文章于 2024-06-05 15:16:09 发布

m0_38116269

最新推荐文章于 2024-06-05 15:16:09 发布

阅读量129

点赞数

分类专栏： caffe 个人学习文章标签： caffe

本文链接：https://blog.csdn.net/m0_38116269/article/details/88980409

版权

个人学习同时被 2 个专栏收录

74 篇文章 1 订阅

订阅专栏

caffe

9 篇文章 0 订阅

订阅专栏

这篇博客记录了在训练resnet-18、resnet-32以及GoogLeNet模型时的前向和反向传播时间，详细对比了不同模型的运算效率。

摘要由CSDN通过智能技术生成

resnet-18

./build/tools/caffe time --model=models/resnet-18/train_val.prototxt -iterations 100 -gpu 0

Result:

I0402 20:25:53.149813 152504 caffe.cpp:408] Average Forward pass: 27.1067 ms.
I0402 20:25:53.149821 152504 caffe.cpp:410] Average Backward pass: 51.3794 ms.
I0402 20:25:53.149834 152504 caffe.cpp:412] Average Forward-Backward: 78.5821 ms.
I0402 20:25:53.149845 152504 caffe.cpp:414] Total Time: 7858.21 ms.
I0402 20:25:53.149853 152504 caffe.cpp:415] *** Benchmark ends ***

resnet-32:

I0402 20:28:40.530372 152672 caffe.cpp:408] Average Forward pass: 66.8848 ms.
I0402 20:28:40.530381 152672 caffe.cpp:410] Average Backward pass: 110.542 ms.
I0402 20:28:40.530391 152672 caffe.cpp:412] Average Forward-Backward: 177.526 ms.
I0402 20:28:40.530400 152672 caffe.cpp:414] Total Time: 17752.6 ms.
I0402 20:28:40.530405 152672 caffe.cpp:415] *** Benchmark ends ***

GoogLenet:

I0402 20:30:18.424845 152775 caffe.cpp:408] Average Forward pass: 36.2463 ms.
I0402 20:30:18.424854 152775 caffe.cpp:410] Average Backward pass: 70.0618 ms.
I0402 20:30:18.424865 152775 caffe.cpp:412] Average Forward-Backward: 106.421 ms.
I0402 20:30:18.424872 152775 caffe.cpp:414] Total Time: 10642.1 ms.
I0402 20:30:18.424878 152775 caffe.cpp:415] *** Benchmark ends ***

4.VggNet(old)

I0402 20:32:56.506058 152895 caffe.cpp:408] Average Forward pass: 108.767 ms.
I0402 20:32:56.506065 152895 caffe.cpp:410] Average Backward pass: 333.067 ms.
I0402 20:32:56.506074 152895 caffe.cpp:412] Average Forward-Backward: 441.922 ms.
I0402 20:32:56.506084 152895 caffe.cpp:414] Total Time: 44192.2 ms.
I0402 20:32:56.506090 152895 caffe.cpp:415] *** Benchmark ends ***