Descriptions
- Offered by: Carnegie Mellon University
- Prerequisites: No strict prerequisites; introductory machine learning and hands-on deep learning training experience are recommended. Familiarity with PyTorch and basic CUDA/GPU concepts is helpful.
- Programming Languages: Python; CUDA and hardware concepts are involved in systems and kernel-level topics.
- Difficulty: πππππ
- Class Hour: 80-120 hours
A systems-focused course on how high-level machine learning models are compiled and executed efficiently on accelerators and distributed infrastructure. Topics include GPU and CUDA programming, ML compilation, graph optimization, distributed training, automatic parallelization, and LLM serving and inference acceleration. Coursework includes weekly paper reviews and a team systems project.
Resources
- Course website: Course website
- Schedule and readings: Schedule and readings
- Lecture slides: Lecture slides
- Logistics and project requirements: Logistics and project requirements
- Preparatory materials: Preparatory materials
- Automated source: cs-self-learning