Project Outcomes

FATES-MLOps project logo

Last update 2026-09-18.

Project Outcomes

The FATES-MLOps project aims to systematically integrate Fairness, Accountability, Transparency, Ethics, and Security principles into MLOps practices. Here are the key outcomes achieved so far:

Scientific Contributions & Academic Dissemination

Publications

See the dedicated page: Publications or our HAL collection.

Conference Presentations

In addition to the conferences where we have published (see above), our team has presented findings at various events:

  • Explain’AI 2025 - 4th Edition (January 28th, 2025)

    • Location: INSA de Strasbourg

    • Slides available: PDF

  • Journées du GdR GPL (June 16th, 2025)

    • Location: Pau, France

    • GT GL&AI slides: PDF

    • GT IE slides: PDF

  • GT IASIV Thematic Day (September 16th, 2025)

    • Location: Central Supélec, Saclay, France

    • Organized by Groupe de Travail sur l’Intelligence Artificielle Soutenable, Intelligible et Vérifiable

Training and Capacity Building

PhD Students

The project is currently supporting multiple PhD students across partner institutions:

Internships

  • Emré Alkan & Kaci Ibazizene - Internship study on FATES properties in MLOps (2025)

Collaborative Network

International Partnerships

The project has strengthened collaborations between:

Research Community Integration

Our team actively participates in:

  • GdR SciLog (Groupe de Recherche Sciences du Logiciel, previously called GPL)

  • GT IASIV (Intelligence Artificielle Soutenable, Intelligible et Vérifiable)

  • GT EXPLICON (Explicabilité et Confiance)

  • RE (Requirements Engineering) community

Resources and Deliverables

Documentation and Materials

Software Tools

As part of our research on MLOps practices and tooling, we have developed several software tools:

  • utrain - A tool for training small language models on local hardware

Link: Documentation

Description: utrain manages the end-to-end training process for small language models on your computer with your GPU. It handles containers, configuration, monitoring, and metrics so you can concentrate on the model. Each run is made of ordered phases (tokenizing, pretraining, etc.) that can be monitored, stopped, restarted, and inspected.

  • naw - Machine learning metrics logging and visualization tool

Link: Documentation

Description: naw logs machine-learning time-series metrics to compressed files on disk and provides a CLI to read them back. It includes a wandb-compatible façade for easy integration with existing code. naw uses efficient compression schemes for different data types and supports live monitoring of runs in progress.

  • colombus - The ML pipelines exploration platform

Link: GitHub Repository

Description: Tool developed at CNRS to explore ML pipelines.

Contact

For more information about project outcomes or collaboration opportunities, please contact: contact@fates-mlops.org