Training Loop Config

class mlip.training.training_loop_config.TrainingLoopConfig(*, num_epochs: Annotated[int, Gt(gt=0)], num_gradient_accumulation_steps: Annotated[int, Gt(gt=0)] = 1, random_seed: int = 42, ema_decay: Annotated[float, Gt(gt=0.0), Le(le=1.0)] = 0.99, use_ema_params_for_eval: bool = True, eval_num_graphs: Annotated[int, FieldInfo(annotation=NoneType, required=True, metadata=[Gt(gt=0)])] | None = None, run_eval_at_start: bool = True)

Pydantic config holding all settings related to the TrainingLoop class.

num_epochs

Number of epoch to run.

Type:

int

num_gradient_accumulation_steps

Number of gradient steps to accumulate before taking an optimizer step. Default is 1.

Type:

int

random_seed

A random seed, by default set to 42.

Type:

int

ema_decay

The EMA decay rate, by default set to 0.99.

Type:

float

use_ema_params_for_eval

Whether to use the EMA parameters for evaluation, set to True by default.

Type:

bool

eval_num_graphs

Number of validation set graphs to evaluate on. By default, this is set to None which means to evaluate on all the available graphs.

Type:

int | None

run_eval_at_start

Whether to run an evaluation on the validation set before we start the first epoch. By default, it is set to True.

Type:

bool

__init__(**data: Any) → None

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.