facebookarchive / caffe2

Caffe2 is a lightweight, modular, and scalable deep learning framework.
https://caffe2.ai
Apache License 2.0
8.42k stars 1.95k forks source link

Reduce the use of memory of network #689

Open lzx1413 opened 7 years ago

lzx1413 commented 7 years ago

When I deploy a caffe2 net on the mobile, is there any means that can reduce the memory usage such as some inplace operators? I have tested that whether the name of input and output of an operator are the same or not, the size of memory the predictor used is the same.thanks

akyrola commented 7 years ago

Use the memonger.py module: memonger.optimize_interference() is probably what you need

mynameischaos commented 7 years ago

Is there any c++ api?