Platforms/Software Sys for LLM

Course Description

AI/ML systems have become very pervasive over the recent years and the energy needs of AI/ML data centers, driven by the use of foundational/large language models (LLMs) as used in Bard, Chat GPT (and others) are a leading contributor to the rising electricity demands of all computing systems in use today. This course will focus on emerging hardware and software solutions for implementing AI/ML systems, particularly those used for supporting LLMs. The software solutions will overview platform and programming environments for current products from Nvidia, Google, AMD, IBM and Intel. The software systems will also examine algorithms, scheduling and mapping techniques for supporting the LLMs on hardware. An introduction to neural networks and their use in LLMs, existing AI hardware and programming environments will be provided by the instructor. This course will then look at emerging research into hardware and software directed at supporting LLMs efficiently from both the energy-usage and performance perspectives. Students are required to read research papers and submit a critical analysis, followed by a detailed, in-person presentation in class. Prerequisite: CS 520 Computer Organization and Architecture or equivalent or CS576 Programming Models for Emerging Platforms. An introductory graduate course on AI or Machine learning is useful, but not mandated. Any exceptions are based on the permission of the instructor. Expected to be offered at least once every two years.

Code

CS 680J

Credits

—

Subject

CS