Great question! Let's break down both of your proposed approaches to move Hydra's hydra.yaml output away from the hardcoded outputs directory—both are totally feasible, and here's how each would work:

Approach 1: Disable Default outputs Directory Output During Initialization #

This is absolutely doable with Hydra's existing configuration options. Hydra lets you control whether it creates the default outputs directory and saves the config there via built-in settings:

• Via config file: Add this to your Hydra config (e.g., config.yaml):

  hydra:
    output_subdir: None
    run:
      dir: .
  

  Setting output_subdir: None tells Hydra not to create the nested timestamped subdirectory under outputs, and run.dir: . sets the base run directory to your current working directory (you can replace . with any path you want if you still want a custom base dir without auto-generated subdirs).

• Via command line: Override these settings on the fly when running your script:

  python your_script.py --hydra.output_subdir=None --hydra.run.dir=/path/to/your/custom/dir
  

  This skips the default outputs directory entirely, so no hydra.yaml will be saved there unless you explicitly configure it.

Approach 2: Add a Hydra Method to Export Config to a Custom Directory #

This is also feasible—either using Hydra's existing APIs to roll your own solution, or proposing an official method addition to the Hydra codebase.

Using Existing APIs (No Need to Modify Hydra) #

You can already access the current Hydra config and save it to any path manually. Here's a quick example:

import hydra
from omegaconf import OmegaConf
from hydra.core.hydra_config import HydraConfig

@hydra.main(config_path="configs", config_name="config")
def app(cfg):
    # Get the full Hydra config object
    hydra_cfg = HydraConfig.get()
    # Define your custom save path
    custom_path = "/path/to/your/custom/hydra.yaml"
    # Serialize and save the config to the custom location
    OmegaConf.save(config=hydra_cfg, f=custom_path)
    # Rest of your application logic...

if __name__ == "__main__":
    app()

This uses OmegaConf (Hydra's underlying config library) to serialize the Hydra config and write it directly to your desired location.

Proposing an Official Method Addition #

If you want to see a dedicated, built-in method (like hydra.save_config_to(path)), this is a valid feature request. Hydra's codebase is designed for extensibility:

• You could contribute a method to the HydraConfig class that wraps the serialization and file-writing logic, making it a clean, reusable API for all users.
• The Hydra community actively accepts feature proposals that improve usability, so this would likely be well-received as a way to simplify config export workflows.

Both approaches solve the problem of moving hydra.yaml away from the default outputs directory—pick the one that best fits your workflow!

内容的提问来源于stack exchange，提问作者Hugh Perkins