Simple type-safe @dataclass CLI with hydra_zen.zen()

[kpe]

(I'm very new to hydra and hydra-zen, and cannot quite wrap my head around it)
I'm trying to create a simple CLI with hydra_zen.zen() (while staying as pythonic as possible) like:

python run.py task=train train.lr=0.01
python run.py task=eval eval.model_path=model.pth

I understand, I could do something like:

from dataclasses import dataclass, field
from hydra_zen import ZenStore, zen

@dataclass
class TrainConfig:
    lr: float = 0.01
    num_epochs: int = 5
    batch_size: int = 32

@dataclass
class EvalConfig:
    model_path: str = "model.pth"
    batch_size: int = 16

@dataclass
class MainConfig:
    task = "train"
    train: TrainConfig = field(default_factory=TrainConfig)
    eval: EvalConfig = field(default_factory=EvalConfig)

# def main(cfg: MainConfig) -> None:
#     print(OmegaConf.to_yaml(cfg))

def main_fn(task: str, train: TrainConfig, eval: EvalConfig) -> None:
    print(OmegaConf.to_yaml(cfg))

store = ZenStore()
store(MainConfig, name="main")

if __name__ == "__main__":
    store.add_to_hydra_store()
    zen(main_fn).hydra_main(
        config_name="main",
        config_path=None,
        version_base="1.3",
    )

to call main_fn() by defining the arguments using a @dataclass , but would it be somehow possible to call the main method directly, which has all the configuration passed as a single cfg: MainConfig argument, while keeping the above CLI interface?
If I would just use zen(main) then the configuration would be expected as cfg.train.lr=0.01 and not train.lr=0.01. Is there some way around this?

[rsokl]

What you are asking for is Hydra's default behavior; just use hydra.main instead of zen.

But note that it is typically better design to have your function deal not with configs but the objects that are instantiated from those configs. That is the main differentiating factor of hydra-zen apps vs Hydra apps.

[kpe]

I see, I guess the following part of the existing documentation explains the point - https://mit-ll-responsible-ai.github.io/hydra-zen/explanation/dont_repeat_yourself.html