We export the CUDA version if SINGA is compiled with CUDA enabled. The cuDNN version is fixed by SINGA and cuDNN is installed from anaconda cloud.
# for SINGA with GPU, e.g. cuda9.0-cudnn7.3.1 export CUDA=9.0
Then, we export a flag DIST to indicate if SINGA is compiled with distributed training enabled.
# to enable distributed training: DIST=ON, otherwise: DIST=OFF export DIST=OFF
We need to export both CUDA and DIST for GPU version. For CPU-only version, we do not export CUDA and DIST.
After exporting the environment variables, we need to add the necessary conda channels
conda config --add channels conda-forge conda config --add channels nusdbsystem
Then, we can execute the following commands to compile SINGA and package it
conda-build . --python 3.6
You will see the package path from the screen output, e.g.,
To clean the cache
conda clean -ay