Which kind of support can you expect from MINERVA?
From rapid technical troubleshooting to sustained engineering collaboration
We know your mileage may vary, so we have developed a support structure that applies to one service levels, depending on the complexity and context of the request.
We defined four types of support, defined by their format, duration and depth. This structure helps the coordination of service delivery, management of resources, and ensure consistent handling of your requests.
Assistance
Short-term, reactive technical help
Assistance is designed to offer quick, targeted support in response to user issues — typically those related to the use of HPC infrastructure or AI tools. This type of support is triggered through the MINERVA ticketing system.
Consulting
Scheduled advisory sessions on planning and theoretical topics
Consulting provides structured guidance to help users plan their work, understand HPC requirements, navigate ethical and regulatory frameworks, or prepare proposals. It is not hands-on engineering work, but rather expert consultation.
workshop
Time-boxed co-working sessions with technical experts
Workshops bring together users and engineers in short, focused sessions to optimize performance, accelerate training workflows, or overcome architectural bottlenecks. These are more involved than consulting, with direct technical collaboration.
Embedded support
Medium- to long-term technical collaboration
This support type involves the integration of a MINERVA engineer into a user project over a longer period. It supports the co-development or scaling of substantial workflows, particularly for advanced AI training or model fine-tuning.
Not all support types are equally suited to all service levels
Their compatibility depends on several factors, including the nature of the task, the time required for delivery, and the availability of resources across MINERVA partners.
A few examples
The following examples illustrate typical use cases encountered by the MINERVA team.
The goal is to provide a reference, showing how thematic challenges translate into practical engagements.
ASSISTANCE
Example with L1 – Porting
A user encounters an error when submitting a training job on an HPC cluster using Slurm. The issue is resolved via ticket-based support by adjusting their job script and environment setup.
Example with L3 – Specialization
A user faces instability when loading a pre-trained model in their fine-tuning script. A quick investigation helps resolve versioning conflicts in the software stack.
consulting
Example with L2 – Scaling
A scientific team has trained an initial model and obtained promising results. They now wish to scale up by increasing dataset size and resolution. A consulting session is held to discuss available HPC resources, best practices in distributed training, considerations for data storage and I/O, and scaling strategies in deep learning (data vs. model vs. pipeline parallelism).
Example with L5 – Ethical
A start-up working on medical AI products requests advice on compliance with the EU AI Act. A consulting session walks through potential regulatory risks and documentation requirements.
workshop
Example with L2 – Optimization
The workshop focuses on helping users reconfigure their training pipelines using mixed precision and data loading optimizations for better performances and on GPU clusters.
Example with L3 – Fine-Tuning
A team working on LLM specialization participates in a two-day workshop with MINERVA engineers to profile a model using PEFT techniques on multi-GPU setups.
embedded support
Example with L1 – Workflow Adaptation
A research group developing an HPC-based AI workflow is assigned a MINERVA engineer for a few weeks to help refactor their codebase, integrate data preprocessing, and validate performance.
Example with L3 – Pre-training
In collaboration with a national institution, a large-scale effort is launched to pre-train a domain-specific LLM. A MINERVA engineer is embedded in the project for a few months to assist with resource acquisition, workflow development, model scaling, and optimization. Acting as a bridge, the engineer transfers expertise from MINERVA’s benchmark data, tools, and HPC best practices into the project team.
Request support sending an email to: support@minerva4ai.eu or fill the form below: