Deep Learning with Keras, Tensorflow, PyTorch, and Horovod on Theta

Presented by: Huihuo Zheng Overview: Datascience modules on Theta • How we built Tensorflow, Pytorch, Keras, and Horovod • Best practices for using the modules / installing other python packages • OMP threading and environmental variables setup (Tensorflow) • Data distribution parallelization with Horovod and Cray ML plugin • Visualization through Tensorboard • Profiling using timeline trace and Vtune