Developer Installation

This guide provides step-by-step instructions for installing the itwinai library for developers.

Cloning the Repository

When cloning the repository, you have to make sure to also clone the submodules. You can do both with the following command:

git clone [--recurse-submodules] git@github.com:interTwin-eu/itwinai.git

Where the --recurse-submodules is an optional flag that allows to pull also git submodules. It is not generally needed.

Setting up the system dependencies

First of all, before installing itwinai and its Python dependencies let’s make sure that the system dependencies such as CUDA drivers, compilers, and MPI libraries, are correctly set up.

Supported OSs are Linux and macOS.

Warning

On high-performance computing (HPC) systems, you must load the appropriate modules before creating or activating your Python virtual environment to ensure compatibility with system libraries.

On HPC systems, it is common to manage dependencies using Environment Modules or Lmod. These tools allow you to dynamically load and unload software modules, such as compilers, CUDA drivers, and MPI libraries. If you are unsure which modules to load for your application, contact your system administrator or refer to your HPC system’s documentation for specific guidance.

Find below the modules you should load on the supercomputers where we tested itwinai, depending on whether you want PyTorch or TensorFlow support:

Modules for the JUWELS system at Juelich Supercomputer (JSC):

ml --force purge
ml Stages/2024 GCC OpenMPI CUDA/12 cuDNN MPI-settings/CUDA
ml Python CMake HDF5 PnetCDF libaio mpi4py

# Now you can create or active the python environment here

ml --force purge
ml Stages/2024 GCC/12.3.0 OpenMPI CUDA/12 MPI-settings/CUDA
ml Python/3.11 HDF5 PnetCDF libaio mpi4py CMake cuDNN/8.9.5.29-CUDA-12

# Now you can create or active the python environment here

Modules for Vega Supercomputer:

ml --force purge
ml Python/3.11.5-GCCcore-13.2.0 CMake/3.24.3-GCCcore-11.3.0 mpi4py OpenMPI CUDA/12.3
ml GCCcore/11.3.0 NCCL cuDNN/8.9.7.29-CUDA-12.3.0 UCX-CUDA/1.15.0-GCCcore-13.2.0-CUDA-12.3.0

# Now you can create or active the python environment here

ml --force purge
ml Python/3.11.5-GCCcore-13.2.0 CMake/3.24.3-GCCcore-11.3.0 mpi4py OpenMPI CUDA/12.3
ml GCCcore/11.3.0 NCCL cuDNN/8.9.7.29-CUDA-12.3.0 UCX-CUDA/1.15.0-GCCcore-13.2.0-CUDA-12.3.0

# Now you can create or active the python environment here

On LUMI, Python virtual environments are not allowed, in favour of containers. Therefore, the software modules are a bit different (WIP).

Creating a Python Virtual Environment

The suggested way of managing Python dependencies, including itwinai, is through Python virtual environments. Creating a virtual environment is allows to isolate dependencies and prevent conflicts with other Python projects.

Beware that some HPC centers advise against using Python virtual environments as they create a large amount of files, which can clog some distributed filesystems. In such situation, you should prefer using containers.

To manage python virtual environments we use UV, which can be installed from this page. Learn more on UV package manager from our UV tutorial

If you don’t already have a virtual environment, you can create one with the following command:

# Remember to load the software modules first (see section above)!

uv venv

# Alternatively to the command above, if you just want to use plain pip instead of UV
python -m venv .venv

Notice that a new directory called .venv is created to contain your virtual environment. Now, you can start your virtual environment with the following command:

# Remember to load the software modules first (see section above)!

source .venv/bin/activate

Installing the itwinai Library as a Developer

In this project, we use uv as a project-wide package manager. Therefore, we suggest that you skim through the uv tutorial before continuing this tutorial.

Optional Dependencies (extras)

The itwinai library has numerous optional dependencies that can be activated when installing with pip through extras:

  • dev: for developers, including libraries for running tests etc.

  • torch: for installation with PyTorch support.

  • tf: for installation with TensorFlow support.

  • tf-cuda: for installation with TensorFlow support with GPU acceleration enabled.

  • docs: for installation of packages required to build the docs locally.

  • hpo: for installation of packages used for hyperparameter optimization (HPO).

You can at any point in time find (or update) the optional dependencies in the pyproject.toml file in the root of the repository.

Installing the itwinai Library

As a developer, you will also install the library using pip (or uv pip if you wish), but the main difference is that you need to install it as editable using the -e flag. Another difference is that you also need the dev extra.

Note

When installing on HPC, it is sometimes an advantage to use the --no-cache-dir option to avoid filling up your ~/.cache directory. Filling up this directory will often lead to you use up your disk quota, especially in terms of inodes.

Below you can find complete commands for installation, depending on if you are installing itwinai with or without GPU (CUDA) support and locally or on HPC:

uv sync --extra torch --extra dev

# Or alternatively, using pip
uv pip install -e ".[torch,dev]"

uv sync --extra torch --extra dev

# Or alternatively, using pip
uv pip install -e ".[torch,dev]" \
    --extra-index-url https://download.pytorch.org/whl/cu121

Note: This is the same as Local (CUDA) but without using the cache directory.

uv sync --extra torch --extra dev

# Or alternatively, using pip
uv pip install -e ".[torch,dev]" \
    --no-cache-dir \
    --extra-index-url https://download.pytorch.org/whl/cu121

Note

If you want to use the Prov4ML logger, you need to install it explicitly since it is only available on GitHub:

For systems with Nvidia GPUs:

uv pip install "prov4ml[nvidia]@git+https://github.com/matbun/ProvML@new-main"

For macOS:

uv pip install "prov4ml[apple]@git+https://github.com/matbun/ProvML@new-main"

Installing Horovod and Microsoft DeepSpeed

If you also want to install Horovod and Microsoft DeepSpeed for distributed ML with PyTorch, then make sure to install them after itwinai. You can choose if you want to do this with or without GPU (CUDA) support:

uv pip install --no-cache-dir --no-build-isolation git+https://github.com/horovod/horovod.git
uv pip install --no-cache-dir --no-build-isolation deepspeed

curl -fsSL https://github.com/interTwin-eu/itwinai/raw/main/env-files/torch/install-horovod-deepspeed-cuda.sh | bash

Warning

Horovod requires CMake>=1.13 and other packages Make sure to have them installed in your environment before proceeding.

Warning

The installation of Horovod and DeepSpeed needs to be executed on a machine/node where GPUs are available. On some HPC systems, such as the JUWELS system on JSC, GPUs are not available on login nodes (the host you connect to when you SSH into the system), only on compute nodes. On the JUWELS system, follow these steps:

  1. Create an interactive session on the develbooster partition (the booster partition does not provide internet access on the compute nodes!):

    salloc --account=intertwin --partition=develbooster --nodes=1 --ntasks-per-node=1 --cpus-per-task=4 --gpus-per-node=1 --time=00:20:00

  2. Load the environment modules needed for the installation, and activate the virtual env

    ml --force purge
ml Stages/2024 GCC OpenMPI CUDA/12 MPI-settings/CUDA Python/3.11.3 HDF5 PnetCDF libaio mpi4py
source .venv/bin/activate

  3. Run the command shown above to install Horovod an DeepSpeed

    srun curl -fsSL https://github.com/interTwin-eu/itwinai/raw/main/env-files/torch/install-horovod-deepspeed-cuda.sh | bash