This grouping is used to capture all the work needed to develop unified recipes for training at scale such as utilities for large scale model training, templates for data engineering and utilities to leverage physics constraints.