hydra-zen 0.3.0rc4

Utilities for making hydra scale to ML workflows

hydra-zen helps you configure your project using the power of Hydra, while enjoying the Zen of Python!

hydra-zen eliminates the boilerplate code that you write to configure, orchestrate, and organize the results of large-scale projects, such as machine learning experiments. It does so by providing Hydra-compatible tools that dynamically generate “structured configurations” of your code, and enables Python-centric workflows for running configured instances of your code.

hydra-zen offers:

• Functions for automatically and dynamically generating structured configs that can be used to fully or partially instantiate objects in your application.

• The ability to launch Hydra jobs, complete with parameter sweeps and multi-run configurations, from within a notebook or any other Python environment.

• Incisive type annotations that provide enriched context about your project’s configurations to IDEs, type checkers, and other tooling.

• Runtime validation of configurations to catch mistakes before your application launches.

• Equal support for both object-oriented libraries (e.g., torch.nn) and functional ones (e.g., jax and numpy).

These functions and capabilities can be used to great effect alongside PyTorch Lightning to design boilerplate-free machine learning projects!