Uses the pytorch docker image to build an extended image with jupyter notebook, and some other python utils
Switch branches/tags
Nothing to show
Clone or download
Pull request Compare This branch is even with gregjohnso:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

PyTorch + Jupyter on GPUs (on aws)

Creates a docker image for running PyTorch on NVIDIA GPUs with Jupyter notebook support, and installs a handful of Python utils.

Note: One may need to change the docker image in the Dockerfile from pytorch-cudnnv6 to something else based on what the PyTorch docker build is named.

This branch is different from master in that it installs the package This provides access to some nice add-ones, including:

  • Table of Contents: Automatically adds table of contents in the sidebar of a Jupyter notebook.
  • Autopep8: Automatic code cleanup for pep8 compliance.
  • table_beautifier: Sorting of Pandas DataFrames printed as tables, etc.
  • VIM binding: vim bindings within Jupyter notebook.

Some of the extensions seem not to work (TODO: figure out why), but a few useful ones do.

Installlation Instructions

  • On an Ubuntu system (e.g. aws) install current nvidia drivers:

    • sudo add-apt-repository ppa:graphics-drivers/ppa
    • sudo apt-get update && sudo apt-get install nvidia-378
  • Install nvida docker (and docker):

    • wget -P /tmp
    • sudo dpkg -i /tmp/nvidia-docker*.deb && rm /tmp/nvidia-docker*.deb
  • Install pytorch on docker: (at the time of writing the dockerfile is named pytorch-cudnnv6)

    • git clone
    • cd pytorch && docker build -t pytorch-cudnnv6 .
  • Clone this repo next to the pytorch one and run the build script

  • To run interactively (maybe start a screen session first):

    • bash

    • To use jupyter: once in the docker container

      • jupyter notebook --allow-root
      • and then the notebook should be available on port 9999 on http