Table des matières

MINERVA Project

Context

MINERVA (European Support Centre for Scalable AI Research and Deployment) is a 36-month European project (2025–2027), funded by the EuroHPC Joint Undertaking under the DIGITAL-EUROHPC-JU-2023-AISC-03 call (Grant Agreement No. 101182737).

Its goal is to facilitate the integration of large-scale artificial intelligence by leveraging Europe’s supercomputers, offering the AI community — researchers, industry, and start-ups — a single entry point to a network of expertise and HPC resources tailored to AI needs.

MINERVA acts as a central hub for large-scale AI on HPC, with activities including:

  • porting AI applications and workflows to HPC environments
  • optimization and scaling
  • pre-training and specialization of open foundation models
  • development of best practices for trustworthy AI (compliance, bias, data management)

By bringing together HPC centers and AI actors from across Europe, MINERVA pools expertise, shares resources, and develops new services that go beyond what any single center could provide.

Main Activities

MINERVA’s work covers several complementary areas, from technical support to training, the production of best-practice guides, and community building for the AI/HPC ecosystem.

AI User Support Services The MINERVA Service Portfolio is structured into five thematic levels:

  • L1 – Porting & adapting AI workflows to HPC: resource access, adaptation to HPC environments, preprocessing, storage, tools (Jupyter, Kubernetes…)
  • L2 – Optimization & scaling: scalability, adaptation to accelerators, optimization, compact model variants
  • L3 – Pre-training of foundation models: integrating diverse datasets, optimization, compression
  • L4 – Specialization & fine-tuning: PEFT/LoRA, RAG, alignment, efficient evaluation and inference
  • L5 – Ethical & responsible AI: dataset evaluation, fair workflows, compliance (AI Act)

These services can be delivered through four types of support:

  • T1 – Assistance: quick remote help to resolve a specific issue
  • T2 – Consulting: scheduled sessions for strategic advice, architecture, compliance, planning
  • T3 – Workshop: intensive co-engineering sessions for optimization, profiling, scalability
  • T4 – Embedded Support: longer-term technical collaboration on ambitious AI projects

Training MINERVA offers training sessions and hands-on workshops at different levels, from initial use of AI on HPC to advanced deployment and optimization techniques. The schedule and resources are available on the Training page.

Benchmarking Benchmarking activities cover AI models, software libraries, infrastructures, and hardware, to guide optimization efforts and compare performance across Europe.

Best-Practice Guides The project produces and disseminates practical guides (deployment, optimization, responsible AI) to help the AI community make effective use of HPC resources.

Community and Ecosystem MINERVA runs a Community Hub and a User Advisory Board to gather user needs, share best practices, and foster cross-border collaboration.

Role of IDRIS

IDRIS (CNRS) leads the support activities (Work Package 2 – MINERVA Services), which include:

  • defining and evolving the service portfolio
  • coordinating responses to user requests
  • contributing to the delivery of the different support types (T1 to T4, depending on context)

In addition, IDRIS contributes to other areas of the project:

  • training and workshops
  • benchmarking campaigns
  • development of best-practice guides
  • community building for AI/HPC

Partners

MINERVA brings together several key HPC centers and AI actors from across Europe, working together to provide large-scale support to the AI community: CINECA (coordination), CNRS/IDRIS, CEA, CINES, BSC, FZJ/JSC, UNIMORE, University of Tübingen, iGenius, Silo AI.

Learn More