Breaking change in cli parsing behavior

[zxiu2049]

from dataclasses import dataclass
from typing import List, Sequence
import nemo_run as run



@dataclass
class Model:
    """Dummy model config"""
    hidden_size: int
    num_layers: int
    activation: str

@dataclass
class Optimizer:
    """Dummy optimizer config"""
    learning_rate: float
    weight_decay: float
    betas: List[float]

@run.cli.factory
@run.autoconvert
def my_model(
    hidden_size: int = 256,
    num_layers: int = 3,
    activation: str = 'relu'
) -> Model:
    """Create a model configuration."""
    return Model(hidden_size=hidden_size, num_layers=num_layers, activation=activation)

@run.cli.factory
@run.autoconvert
def my_model_2(
    hidden_size: int = 512,
    num_layers: int = 3,
    activation: str = 'relu'
) -> Model:
    """Create a model configuration."""
    return Model(hidden_size=hidden_size, num_layers=num_layers, activation=activation)

@run.cli.factory
def my_optimizer(
    learning_rate: float = 0.001,
    weight_decay: float = 1e-5,
    betas: Sequence[float] = (0.9, 0.999,)
) -> run.Config[Optimizer]:
    """Create an optimizer configuration."""
    return run.Config(Optimizer, learning_rate=learning_rate, weight_decay=weight_decay, betas=list(betas))

@run.cli.entrypoint
def train_model(
    model: Model = my_model(),
    optimizer: Optimizer = my_optimizer(),
    epochs: int = 10,
    batch_size: int = 32
) -> None:
    """
    Train a model using the specified configuration.

    Args:
        model: Configuration for the model.
        optimizer: Configuration for the optimizer.
        epochs: Number of training epochs. Defaults to 10.
        batch_size: Batch size for training. Defaults to 32.
    """
    print(f"Training model with the following configuration:")
    print(f"Model: {model}")
    print(f"Optimizer: {optimizer}")
    print(f"Epochs: {epochs}")
    print(f"Batch size: {batch_size}")

    # Simulating model training
    for epoch in range(epochs):
        print(f"Epoch {epoch + 1}/{epochs}")

    print("Training completed!")

if __name__ == "__main__":
    run.cli.main(train_model)

In this example when I run

python example.py model=my_model_2 model.num_layers=10

The expected behavior/old behavior is that the cli args are parsed properly. But on main(I don't know exactly when this is broken) getting

Resolved Arguments
┏━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Argument Name        ┃ Resolved Value                                               ┃
┡━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ batch_size           │ 32                                                           │
│ epochs               │ 10                                                           │
│ model                │ Model(num_layers=10)                                         │
│ optimizer            │ Optimizer(learning_rate=0.001, weight_decay=1e-05,           │
│                      │ betas=[0.9, 0.999])                                          │
└──────────────────────┴──────────────────────────────────────────────────────────────┘
Continue? [y/N]: y
Launching train_model...
Unexpected error: Model.__init__() missing 2 required positional arguments: 'hidden_size' and 'activation'

[marcromeyn]

Thanks for reporting this! I put in a small fix for this: #235.