Training in the Use of GPU Computing Resources

by the Computing Center of IN2P3-CNRS & HUMA-NUM

Location: CC-IN2P3, 21 avenue Pierre de Coubertin, 69100 Villeurbanne
Dates: Thursday 27 and Friday 28 November 2025, from 8:30 a.m. to 5:00 p.m.
Organizers: Sébastien Gadrat (CC-IN2P3) and Gérald Foliot (IR* Huma-Num)
Trainers:

Bernard Chambon, Guillaume Cochard, Bertrand Rigaud, Sybille Voisin (CC-IN2P3)
Stéphane Pouyllau, Léa Maronet, Adam Faci, and Emile Provendier (IR* Huma-Num)

Format: In-person only. The training is accessible to people with reduced mobility.
Audience: This training is intended for users of Huma-Num services and aims to provide essential information for using GPU resources.

Context

Between 2024 and 2025, in partnership with CC-IN2P3, 68 new GPUs were acquired to support the use of AI and, more specifically, generative AI (GenAI). These GPUs are accessible through various environments: notebook-based prototyping platforms, compute farms, active LLM servers, visual computing platforms, and web application development environments (RAG, etc.).

The purpose of this training is twofold: first, to introduce participants to each of these environments, and second, to help users position themselves between performing their own AI computations and using preconfigured services.

For example, when using an LLM server such as Ollama, one can perform local computations, use a raw LLM server, access it within a notebook, or deploy it via batch jobs—for instance, to process a corpus of 300,000 documents.

Although the training focuses primarily on AI and GenAI applications, it is also open to participants with other GPU computing needs.

Program

Thursday, 27 November 2025

Introduction

Presentation of CC-IN2P3
Overview of the GPU investment by IR* Huma-Num
Account creation and management

Jupyter Notebook Platform

Platform overview
Creating and using custom kernels
Example: GPU-based classification with MNIST data

Slurm Compute Farm

Introduction to Slurm and specific configurations
Submitting a GPU job (H100 96GB NVL)
ARM partition
Job monitoring and results retrieval

Friday, 28 November 2025

AI for the Humanities and Social Sciences (HSS)

Introduction: “AI and GenAI in the Humanities and Social Sciences: how HSS needs can inform the challenges faced by data and computing centers.”
Limitations and implications of GenAI: hallucinations, ethics, and research impact
Building a conversational agent on research data using an LLM and Retrieval-Augmented Generation (RAG) tools
Using AI to identify and classify data: example of automated image processing with Arkindex and other methods
Additional GPU use cases: LLM servers, web applications

Working Environment

The first half of the training will be devoted to theoretical instruction (access and resource usage), while the second half will focus on hands-on exercises using the Slurm compute farm, the Jupyter notebook platform, and ready-to-use applications.

Access will be provided via a dedicated login created for the training; credentials will be distributed on the first day.

For consistency, please do not use your personal CC account during the training. You will, however, be able to save your notebooks and any data produced over the two days.

You will need a laptop with:

a terminal for SSH connections (command-line access), and
a web browser for accessing the Jupyter notebook platform.

A data center tour is scheduled for Thursday at 1:30 p.m.

Practical Information

Registration
Registration is free but mandatory, subject to seat availability.

Please double-check your email address during registration, as all logistical details (including connection information) will be sent via email.

This training is exclusively reserved for users of IR* Huma-Num services, CC-IN2P3 users, and academic or CNRS partner institutions. Registrations not meeting these criteria may be declined.

Access to CC-IN2P3
Lunch will be provided by IR* Huma-Num.

For additional information, please contact the organizers at:
gerald.foliot@huma-num.fr