mit-ll-responsible-ai/hydra-zen

Create powerful Hydra applications without the yaml files and boilerplate code.

What it solves

hydra-zen is designed to simplify the creation of configurable, repeatable, and scalable workflows. It specifically addresses the pain point of managing complex configuration files (like hand-written YAML files) in Hydra projects, which can become cumbersome and tedious to maintain.

How it works

The library provides functions that dynamically and automatically generate dataclass-based configurations. It also includes a custom config-store API and a task-function wrapper to reduce the amount of Hydra-specific boilerplate code required in a project.

Who it’s for

It is intended for developers and researchers writing either research-grade or production-grade code, particularly those using PyTorch Lightning for machine learning projects.

Highlights

  • Eliminates YAML: Removes the need for hand-written YAML configuration files.
  • Configurable: Allows all aspects of the code to be managed from a single interface (command line or Python function).
  • Repeatable: Automatically saves the full configuration alongside results for self-documenting runs.
  • Scalable: Supports launching multiple runs locally or across cluster nodes.