Skip to content

Latest commit

 

History

History
32 lines (23 loc) · 1.37 KB

installation.rst

File metadata and controls

32 lines (23 loc) · 1.37 KB
Raw

Installation

Prerequisites

RaNNC works only with CUDA devices (CPU only/TPU environments are not supported). RaNNC requires the following libraries and tools at runtime.

  • CUDA: A CUDA runtime must be available at the runtime environment. Currently RaNNC is tested with CUDA 10.2 and 11.0.
  • NCCL: NCCL (Version >= 2.7.3 is required) must be available at the runtime environment. RaNNC uses NCCL both for allreduce and P2P communications.
  • MPI: A program using RaNNC must be launched with MPI. MPI libraries must also be available at runtime. RaNNC is tested with OpenMPI v4.0.5.
  • libstd++: libstd++ must support GLIBCXX_3.4.21 to use the distributed pip packages (The packages are built with gcc 5.4.0).

Installation

This version of RaNNC requires PyTorch v1.8.0. pip packages for linux_x86_64 are available from the following links.

You can create a new conda environment and install RaNNC by the following commands. Set a CUDA version available in your environment.

conda create -n rannc python=3.8
conda activate rannc
conda install pytorch==1.8.0 cudatoolkit=10.2 -c pytorch
pip install pyrannc-0.5-cp38-cp38m-linux_x86_64.whl