LLM Supervised Fine-Tuning for SMEs: From Open Models to In-House Solutions
Workshop Details
Practical Information
A laptop. All required software, datasets, and scripts are provided via the workshop repository. Compute credit for GPU access is kindly provided by Exoscale.
Basic Python programming skills and a general familiarity with how LLMs work, such as understanding what prompts, tokens, inference, and GPU acceleration roughly mean.
About this Workshop
Large Language Models (LLMs) are transforming how information is accessed, processed, and delivered, yet many small and medium-sized enterprises (SMEs) still rely on external API providers, which limits their ability to ensure model and data sovereignty and to shape AI-driven solutions around the concrete needs of their own users. This workshop introduces participants to supervised fine-tuning (SFT) as a practical and accessible way to adapt open-weights LLMs. It is designed for practitioners in SMEs who sit between IT and AI roles and who need hands-on methods to bring LLM customization in-house using fully open-source tools. Through three applied use cases, participants will learn the complete SFT workflow: preparing and structuring domain-specific datasets, designing prompts and templates, optimizing training within available hardware, monitoring progress, evaluating model performance, scaling to multi-GPU setups, preparing a fine-tuned model for deployment, and exposing it through an interactive interface. By the end, participants will be ready to develop and operate customized LLMs that align with their organizational objectives and deliver meaningful value to their end users.
Speakers & Organizers
Andrei Coman
Postdoctoral Researcher
AISLab at HES-SO Valais-Wallis
Andrei Coman is a Postdoctoral Researcher in the AISLab at HES-SO Valais-Wallis. Before joining AISLab, Andrei completed his PhD at the Idiap Research Institute and the Ecole Polytechnique Fédérale de Lausanne (EPFL), where his work focused on deep learning architectures for natural language processing, particularly at the intersection of text and graph representation learning. He also gained experience as an Applied Scientist Intern at Amazon AGI, contributing to retrieval-augmented generation systems and contextual reward modeling. Andrei brings applied research experience and hands-on exposure to real-world challenges, working closely with researchers, engineers, and practitioners across multiple projects and institutions. He values open, respectful, and collaborative work environments and believes that progress in science and technology is best achieved through kindness, curiosity, and shared effort.
Pamela Delgado
Professor
IICT at HEIG-VD/HES-SO
Pamela Delgado is a professor at the IICT at HEIG-VD/HES-SO. Before joining HES-SO, she worked at the SDSC and obtained her PhD from EPFL. Her research focuses on efficiently managing large-scale/limited GPU resources at the intersection between systems and machine learning with a focus on sustainability.