Hydra 1.2 'link' Guide

If you have ever tried to manage a massive Python configuration file full of nested dictionaries, you know the pain. That is why the open-source community fell in love with (from Facebook Research). It allows you to compose dynamic configurations from multiple files and override anything from the command line.

This change allows for better type checking and allows you to run Hydra inside Jupyter Notebooks (finally!) without weird hacks. Yes, but carefully. If you are starting a new project today, use Hydra 1.2 . The new composition rules and Jupyter support are worth it. hydra 1.2

defaults: - storage: aws - optional region: ${storage.region} Hydra was notorious for adding 200–400ms to script startup time because it parsed every @dataclass and OmegaConf structure recursively. For long-running training jobs, this didn't matter. For serverless functions or CLIs? It hurt. If you have ever tried to manage a

If you are on a legacy pipeline of 10,000+ lines of configs, pin your version to hydra-core==1.1.2 for now, but plan the migration. The deprecation of hydra.main means you will need to refactor your entry point logic. This change allows for better type checking and

pip install hydra-core --upgrade Happy composing! Let us know in the comments if you have found the 1.2 resolver syntax tricky—I will be writing a deep dive on that next week.

# Old (Hydra 1.1) @hydra.main(config_path="conf", config_name="config") def main(cfg): ... def main(): cfg = hydra.initialize_and_run(config_path="conf", config_name="config", task_function=my_task)