CS7670: Large Language Model Systems (Seminar)
Course information
- Course Number: CS 7670
- Lectures: Tue/Fri 9:50AM - 11:30AM
- Room: Ryder Hall 156
- Instructor: Cheng Tan
Why this seminar?
The rapid advancement of Large Language Models (LLMs) has revolutionized artificial intelligence, but their growing scale and complexity have created unprecedented systems challenges. Modern LLM systems must address massive computational requirements, complex distributed training architectures, and sophisticated inference pipelines that balance latency, throughput, and cost. These systems span hardware acceleration, distributed computing frameworks, memory optimization techniques, and novel serving infrastructures. As models continue to scale beyond trillions of parameters while applications demand real-time performance, understanding the systems that enable LLMs becomes increasingly critical. This seminar explores cutting-edge research addressing these challenges, examining how innovations in systems design are not merely supporting LLM development but actively shaping their capabilities, efficiency, and practical applications across industries.
Prerequisites and background
- have a basic understanding of machine learning systems (such as TensorFlow or PyTorch)
- have a basic understanding of computer systems (at the level of CS5600 or equivalent)
- be able to read papers
- be familiar with Python programming
The course
- The course introduces the basics of LLM systems during the first four weeks.
- It then examines the current state of LLM systems, focusing on training, serving, and optimization.
- The course is heavily discussion-based, with frequent guest lectures and student-led presentations.
- Students are expected to actively engage by forming and presenting their own opinions on the design and, more importantly, the limitations of today’s LLM systems.