autocast.utils.optimizer

Optimizer configuration utilities.

get_optimizer_config(learning_rate=0.0001, optimizer='adam', scheduler=None, **kwargs)

Create an optimizer configuration dict.

Convenience function for quickly creating optimizer configs in notebooks and scripts without needing full Hydra configuration.

Parameters:

• learning_rate (float) – Learning rate for the optimizer. Default 1e-4.

• optimizer (str) – Optimizer name (‘adam’, ‘adamw’, ‘sgd’). Default ‘adam’.

• scheduler (str | None) – Optional scheduler name (‘cosine’, ‘cosine_with_restarts’, ‘step’, ‘plateau’). Default None (no scheduler).

• **kwargs (Any) – Additional optimizer parameters (e.g., betas, weight_decay, step_size, gamma).

Returns:

Dict containing optimizer configuration compatible with OptimizerMixin.

Return type:

dict[str, Any]