Tamp is an open source C++ library for reducing the space and time costs of deep neural networks, with the goal of making them more applicable to devices with limited resources.
Getting Started with TensorFlow
We recommend an virtualenv installation of TensorFlow. Detail tutorial can be found here
After successfully installed tensorflow, activate virtualenv by:
$ source ~/tensorflow/bin/activate # If using bash
$ source ~/tensorflow/bin/activate.csh # If using csh
After that, using build script:
$ ./build_tensorflow.sh # If linking against Intel MKL
$ ./build_tensorflow-cblas.sh # If linking against CBLAS/ATLAS
Getting Started with Caffe
Using the our forked version of Caffe is required since the vanilla Caffe does not have an interface for Tamp-styled plugins in original version.
First, follow the instructions to build the Caffe environment. Some required protocol buffer headers are generated during this build process for Tamp integration.
Next, using build script, run:
Once this is done, turn on Tamp by setting USE_TAMP in Caffe's Makefile.config to include the Tamp library and rebuild Caffe.