Last updated: | Permalink
Lecture 7: ML systems
Learning objectives:
In this lecture, you will learn:
- Lec7a: how the offline Belady caching policy works and why it’s optimal
- Lec7a: the insight that SHADE builds on
- Lec7b: know the concepts of model pruning and quantization
- Lec7b: understand how ELF (exponent-less floating-point compression) works
Lecture slides
- Lec7a: Deep learning caching: slides pdf
- Lec7b: Model compression: slides pdf
Readings
- Lec7a: SHADE paper (optional)
- Lec7b: ELF paper (Section 1-4, Section 6.1, optional)