Skip to content

Latest commit

 

History

History
65 lines (45 loc) · 2.1 KB

installation_guide.md

File metadata and controls

65 lines (45 loc) · 2.1 KB

GPU Cluster Setup Guide for Pytorch

Target environment

  • Ubuntu 18.04
  • Python 3.6
  • PyTorch v1.1
  • CUDA 10.0

Install dependencies

$ sudo apt install build-essential python3-pip

Install CUDA 10.0

CAVEATS: Pytorch v1.1 does not support CUDA 10.1 yet (Only CUDA 9.0 and 10.0 are supported)

You can download the CUDA 10.0 installer for Ubuntu 18.04 here. In this document, we are using the runfile version installer.

$ wget https://developer.nvidia.com/compute/cuda/10.0/Prod/local_installers/cuda_10.0.130_410.48_linux
$ sudo sh cuda_10.0.130_410.48_linux

Then, follow the command-line prompts. NVIDIA driver should be installed if not installed before.

By default, CUDA library is installed at /usr/local/cuda.

Add CUDA libraries to LD_LIBRARY_PATH

  • Option 1: User-level
    • Put the following line to ~/.bashrc or ~/.zshrc: export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64
  • Option 2: System-level
    • Create a new file cuda.conf at /etc/ld.so.conf.d/ and put the following line to the file: /usr/local/cuda/lib64

Check GPU status

$ nvidia-smi

Install PyTorch

On Ubuntu 18.04, the default version of Python 3 is 3.6. PyTorch has pre-built wheels for Python 3.6 and CUDA 10.0.

We are installing PyTorch via pip. You can check the PyTorch website for other options.

$ pip3 install https://download.pytorch.org/whl/cu100/torch-1.1.0-cp36-cp36m-linux_x86_64.whl
$ pip3 install https://download.pytorch.org/whl/cu100/torchvision-0.3.0-cp36-cp36m-linux_x86_64.whl

Now, PyTorch v1.1 is installed. We can check the installed PyTorch version as follows.

$ python3 -c "import torch; print(torch.__version__)"

Miscellaneous

The federated framework uses pre-processed federated datasets that are formatted in HDF5. To read HDF5 in Python, the following module should be installed.

$ pip3 install h5py