Welcome to itwinai

Accelerate AI & ML workflows for Scientific Digital Twins.

itwinai streamlines distributed training, hyperparameter optimization, logging, and modular workflows, so you can focus on science, not plumbing.

Features

• 🚀 Seamless Scaling: Run training and inference on HPC clusters or cloud with a single CLI command.

• 🔍 Effortless Logging: Built-in support for MLflow, Weights & Biases, TensorBoard, and more.

• 🧩 Modular Workflows: Define reusable pipelines for end-to-end experiment management.

• 🤖 HPO Made Easy: Native hyperparameter optimization with minimal configuration.

• 🔌 Extensible Plugins: Add custom integrations or contribute new features.

Quick Start

# Install via pip
pip install itwinai

# Launch a complete workflow with SLURM integration using the MNIST example
itwinai run -c https://raw.githubusercontent.com/interTwin-eu/itwinai/refs/heads/main/use-cases/mnist/torch/run-example.yaml

# View logs in MLflow
itwinai mlflow-ui --path mllogs/mlflow

🚀 Begin Here

• User Installation (for non-developers)

• Developer Installation

• Submitting jobs to SLURM on HPC

• Using itwinai Container Images

🛠️ Core Guides

• Training a Neural Network

• Logging and Tracking ML workflows

• Defining machine learning workflows

• Hyperparameter Optimization

🎓 Tutorials

• Writing Configuration Files for itwinai

• Distributed Training

• Hyper-parameter Optimization

• ML Workflows

• Code Profiling and Optimization

📚 Use Cases & 🧩 Plugins

• MNIST — A Toy Use Case Example

• Drought Early Warning in the Alps (EURAC)

• Fast particle detector simulation (CERN)

• Writing Plugins for itwinai

• Current List of itwinai Plugins

For the full list of scientific use cases refer to the navigation side bar.

⚡ API Reference

• CLI Reference

• Python SDK

Integration with EuroHPC centers

Our code has been run and tested on the following EuroHPC systems:

• LUMI

• JSC

• Vega

• Deucalion (coming soon)

Community & Support

• Join our Discord

• GitHub Repository

• Contributors

• interTwin Project

itwinai is an open-source Python library primarily developed by CERN, in collaboration with Forschungszentrum Jülich (FZJ). As the primary contributor, CERN will retain administrative rights to the repository during and after the interTwin project, except in cases where CERN is unable to maintain it.

How to contribute

Want to help improve itwinai? Here are a few good ways to get involved:

• Report a bug / request a feature: open a GitHub issue.

• Contribute code or docs: fork the repository and submit a pull request.

• Ask questions or float ideas: start a GitHub discussion or join us on Discord.

Indices and Tables

• Index

• Module Index

• Search Page

Citation

If you use itwinai in your research, please cite:

Bunino et al., (2026). itwinai: A Python Toolkit for Scalable Scientific
Machine Learning on HPC Systems. Journal of Open Source Software, 11(117),
9409. https://doi.org/10.21105/joss.09409

BibTeX:

@article{Bunino2026,
  doi = {10.21105/joss.09409},
  url = {https://doi.org/10.21105/joss.09409},
  year = {2026},
  publisher = {The Open Journal},
  volume = {11},
  number = {117},
  pages = {9409},
  author = {Bunino, Matteo and Sæther, Jarl and Eickhoff, Linus and Lappe, Anna and
            Tsolaki, Kalliopi and Verder, Killian and Mutegeki, Henry and Machacek,
            Roman and Girone, Maria and Krochak, Oleksandr and Rüttgers, Mario and
            Sarma, Rakesh and Lintermann, Andreas},
  title = {itwinai: A Python Toolkit for Scalable Scientific Machine Learning on HPC Systems},
  journal = {Journal of Open Source Software}
}