Skip to content

Latest commit

 

History

History
59 lines (45 loc) · 2.4 KB

Colab_instructions.md

File metadata and controls

59 lines (45 loc) · 2.4 KB
Raw

Open In Colab

Instructions for Google Colab (experimental)

Google Colab provides limited free access to an Nvidia GPU, but Julia and the required packages have to be installed manually in the Colab runtime. Note that the Julia ecosystem is not officially supported, so consider this an experimental setup.

Follow these steps each time you start a new runtime:

  1. Connect to a GPU runtime:
    • In the Google Colab interface, click the arrow (▼) next to Connect and select Change runtime type
    • Under Hardware accelerator, select some option with "GPU" in its name
    • Save the runtime settings and click Connect
  2. Create a new code cell, copy the script below and run it. Wait a few minutes for the Julia install to finish.
%%shell
set -e

JULIA_VERSION="1.9.4"

if [ -z `which julia` ]; then
    JULIA_VER=`cut -d '.' -f -2 <<< "$JULIA_VERSION"`
    echo "Installing Julia $JULIA_VERSION on the current Colab Runtime..."
    BASE_URL="https://julialang-s3.julialang.org/bin/linux/x64"
    URL="$BASE_URL/$JULIA_VER/julia-$JULIA_VERSION-linux-x86_64.tar.gz"
    wget -nv $URL -O /tmp/julia.tar.gz
    tar -x -f /tmp/julia.tar.gz -C /usr/local --strip-components 1
    rm /tmp/julia.tar.gz
    
    echo "Installing IJulia package..."
    julia -e 'using Pkg; pkg"add IJulia; precompile;"' &> /dev/null
    
    echo "Installing IJulia kernel..."
    julia -e 'using IJulia; IJulia.installkernel("julia")'
    KERNEL_DIR=`julia -e "using IJulia; print(IJulia.kerneldir())"`
    KERNEL_NAME=`ls -d "$KERNEL_DIR"/julia*`
    mv -f $KERNEL_NAME "$KERNEL_DIR"/julia
fi

git clone https://github.com/sfalmo/NeuralDFT-Tutorial.git
cp NeuralDFT-Tutorial/*.jl NeuralDFT-Tutorial/*.toml .

echo "Done."
  1. Switch to the Julia kernel:
    • Click the arrow (▼) next to Connect and select Change runtime type
    • Under Runtime type, select the option "julia 1.9.4" (do not change the Hardware accelerator which you have chosen in step 1)
  2. Install the required Julia packages: create a new cell, copy the following code and run it. This will take a few minutes.
import Pkg
Pkg.activate(".")
Pkg.instantiate()

You can now start the tutorial. As all required packages have just been installed, the GPU is automatically used where applicable.