Installation#

Supported Environments#

These are the current environments supported by GiGL

Python

Mac (Arm64) CPU

Linux CPU

Linux CUDA

PyTorch

PyG

3.9

Partial Support

Supported

12.1

2.5

2.5

Install Prerequisites - setting up your dev machine#

Below we provide two ways to bootstrap an environment for using and/or developing GiGL

(Recommended) Developing/experimenting on a GCP cloud instance.

  1. Create dev instance We will need to create a GCP instance and setup needed pre-requisites to install and use GiGL.

You can use our create_dev_instance.py script to automatically create an instance for you:

  python <(curl -s https://raw.githubusercontent.com/Snapchat/GiGL/refs/heads/main/scripts/create_dev_instance.py)

Next, ssh into your instance. It might ask you to install gpu drivers, follow instructions and do so.

  1. Install some pre-reqs on your instance. The script below tries to automate installation of the following pre-reqs: make, unzip, qemu-user-static, docker, docker-buildx, mamba/conda

  bash -c "$(curl -s https://raw.githubusercontent.com/Snapchat/GiGL/refs/heads/main/scripts/scripts/startup_dev_instance.sh)"

  1. Once you are done, make sure to restart the instance. You may also need to navigate to the GCP compute instance UI, and under the Observability tab of your instance click the “Install OPS Agent” button under the GPU metrics to ensure the GPU metrics are also being reported.

Next, Follow instructions to install GiGL

Manual Setup

  1. If on MAC, Install Homebrew.

  2. Install Conda:

  3. Install Docker and the relevant buildx drivers (if using old versions of docker):

    Once installed, ensure you can run multiarch docker builds by running following command:

    Linux:

    docker buildx create --driver=docker-container --use
sudo apt-get install qemu-user-static
docker run --rm --privileged multiarch/qemu-user-static --reset -p yes

    Mac:

    docker buildx create --driver=docker-container --use
brew install qemu
docker run --rm --privileged multiarch/qemu-user-static --reset -p yes

  4. Ensure you have make installed.

    make --version

    Install make on Linux:

    apt-get update && apt-get upgrade -y && apt-get install -y cmake

    Install make on MAC:

    brew install make

    Subsequently, you should be able to use gmake in all places where we use make since brew formula has installed GNU “make” as “gmake”. See: https://formulae.brew.sh/formula/make

  5. Follow the glcoud cli install instructions

  6. Then, setup your gcloud environment:

```bash
gcloud init # setup glcoud CLI
gcloud auth application-default login # Auth gcloud cli
gcloud auth configure-docker us-central1-docker.pkg.dev # Setup docker auth for GiGL images.
```

Install GiGL#

Install Wheel#

  1. Create a python environment

conda create -y --override-channels --channel conda-forge --name gigl python=3.9
conda activate gigl

  1. Install GiGL

Install GiGL + necessary tooling for Torch 2.5 + Cuda12.1#

pip install "gigl[torch25-cuda-121,transform]==0.0.9" \
  --index-url=https://us-central1-python.pkg.dev/external-snap-ci-github-gigl/gigl/simple/ \
  --extra-index-url=https://pypi.org/simple

Currently, building/using wheels for GLT is error prone, thus we opt to install from source every time. Run post-install script to setup GLT dependency:

gigl-post-install

Install GiGL + necessary tooling for Torch 2.5 + CPU#

pip install "gigl[torch25-cpu,transform]==0.0.9" \
  --index-url=https://us-central1-python.pkg.dev/external-snap-ci-github-gigl/gigl/simple/ \
  --extra-index-url=https://pypi.org/simple

Currently, building/using wheels for GLT is error prone, thus we opt to install from source every time. Run post-install script to setup GLT dependency:

gigl-post-install

Install from source#

There are various ways to use GiGL. Firstly, clone the repo locally.

git clone https://github.com/Snapchat/GiGL.git

From the root directory:

make initialize_environment
conda activate gnn

This creates a Python 3.9 environment with some basic utilities. Next, to install all user dependencies. Note: The command below will try its best ot infer your environment and install necessary reqs i.e. if CUDA is available it will try to install the necessary gpu deps, otherwise it will install cpu deps.

make install_deps

If you instead want to contribute and/or extend GiGL. You can install the developer deps which includes some extra tooling useful for contributions:

make install_dev_deps