Section 1 – Data Parallelism
In this section, you will understand why data parallelism is needed and how it works. You will implement data-parallel training and serving pipelines and learn advanced techniques for further speed-ups.
This section comprises the following chapters:
- Chapter 1, Splitting Input Data
- Chapter 2, Parameter Server and All-Reduce
- Chapter 3, Building a Data Parallel Training and Serving Pipeline
- Chapter 4, Bottlenecks and Solutions