This course introduces machine learning on the DelftBlue supercomputer, blending theory with hands-on practice. Participants will learn to manage dependencies, handle large datasets, optimize GPU performance, and scale deep learning models using PyTorch.
At the end of the day, you should be able to
– Understand the basic setup of a supercomputer and identify bottlenecks in ML applications.
– Run PyTorch examples on CPUs and GPUs on DelftBlue, assessing performance and resource requirements.
– Optimize workflows to efficiently utilize a single GPU, including LLM examples.
– Scale a deep learning workflow for distributed training across GPUs.
## Prerequisites
– Python
– Command line and DelftBlue basics
– Experience with machine learning workflows (e.g., PyTorch)
## Content level
– 30% Beginner (Connecting to DelftBlue, dependency management, basics of Slurm)
– 50% Intermediate (Handling datasets, optimizing ML workflows)
– 20% Advanced (Large models on small GPUs, scaling with Fabric)
One day course
30 max. participants
Teachers:
A. Ahmed
S. Wacker
Costs:
€100,-
€25,- for BSc and MSc students.
including lunch and course materials, free for DCSE members.
Location:
Penguinlab, EWI B36.HB.2.130
Prerequisites:
Command line and DelftBlue basics
Python
Basic knowledge of ML algorithms
Note:
This course is organized in the computer lab of the Mathematics department. Exercises will be carried out on DelftBlue.
You can either bring your own laptop or use one of the lab computers as a terminal.