/TSplit

Primary LanguagePythonApache License 2.0Apache-2.0

Athena

A deeplearning framework for 10 billlion parameters DL model.

At present, Tenbillionflow supports for DNN training for both CPU and GPU with manually operators on numpy and CUDA C++ implemantation.

Installation

  1. Clone the respository.
  2. Edit the athena.exp file and set the environment path for python.
source athena.exp
  1. Compile each system by make all For standalone development:
make clean
make athena version=cpu -j 32
make athena version=gpu -j 32
  1. Install graphviz to support graph board visualization
sudo apt-get install graphviz
sudo pip install graphviz
  1. Install pynvml to support vdnn scheduler policy.
sudo pip install nvidia-ml-py

For mkl support

make clean
make athena version=mkl -j 32
cd tests
python mnist_mlp_no_bias.py

To compare with numpy, rebuild with cpu version and run the same test script.