HPC Training Series - Course 9 "Running LLMs on HPC: Transformers, Inference & Deployment"

Friday 17 Jan 2025, 10:00 → 14:30 Europe/Athens

EuroCC@Greece in collaboration with Smart Attica EDIH, announced  the 9th Course of HPC Training Series with the subject "Running LLMs on HPC: Transformers, Inference & Deployment", that took place online on January 17th, 2025. 

Date: January 17th, 2025, at 10:00 EET 

Location: Online via Zoom

Presentation Languages: Greek & English

Audience:

• Data scientists and machine learning engineers.  
• NLP researchers and practitioners.  
• HPC system administrators and engineers.  
• Developers exploring Hugging Face Transformers and RAG.  
• Academic researchers working on language modeling projects.  
• Professionals interested in training or deploying LLMs on HPC.  
• Organizations planning to adopt HPC for AI workloads.

Description: This course focused on Large Language Models running on High-Performance Computing systems. Participants gained a foundational understanding of the Hugging Face Transformers library, embeddings’ models, and of Retrieval-Augmented Generation. They discovered how to effectively set up an inference server on HPC systems as well as a deployment process and limitations. Training of the Greek LLM Meltemi was also presented. This seminar included hands-on sessions where users were be able to run the provided code.

Learning Objectives:

• Hands-on experience on Hugging Face Transformers 
• Set up and troubleshoot LLM inference servers on HPC systems.  
• Explore LLM deployment on HPC, including limitations and applications.  
• Learn the training process of the Greek LLM Meltemi.  
• Understand capacity and scaling challenges in LLM deployment.  
• Experiment with real-world applications of LLMs.

Prerequisites:

• Basic understanding of machine learning and neural networks,
• Knowledge of Python programming,
• Basic command-line and Linux skills.

Note: Please enter your institutional/corporate email when registering.

 

Read the Privacy Statement here.pdf

10:00 → 10:30 Setting up the hands-on environment (local or cloud-based/ HPC), issues solving

10:30 → 10:40 The Greek Competence Center for HPC & AI

Speaker: Mr Ilias Hatzakis (GRNET)

10:40 → 12:00 Introduction to Hugging Face Transformers, embeddings’ models, and basics of Retrieval-Augmented Generation

Speaker: Dr Nikos Bakas (GRNET)

12:00 → 12:45 Using LLMs on “Aristotelis” HPC infrastructure: deployment, experimentation, capacity and limitations, applications

Speaker: Dr George Vlahavas (AUTH)

12:45 → 13:00 Break

13:00 → 13:45 Inference of Hugging Face’s Pre-Trained LLMs on HPC Systems

Speaker: Dr Marco Magliulo (LuxProvide / MeluXina)

13:45 → 14:15 Training the Greek LLM Meltemi

Speaker: Dr Vassilis Katsouros (Athena RC)

14:15 → 14:30 Wrap up / Q&A / Discussion