Skip to content
GitHub
View on GitHub

TrainingGroup

A base TrainConfig plus a grid of field overrides to expand.

from modal_training_gym.common.training_group import TrainingGroup

A base TrainConfig plus a grid of field overrides to expand.

TrainingGroup(base, grid=None, name=None, merge_model_recipe=None)
ParameterTypeDefaultDescription
baseTrainConfigrequired
griddict[str, list[Any]] | NoneNone
namestr | NoneNone
merge_model_recipebool | NoneNone

get_train_configs(self) -> 'list[TrainConfig]'

Section titled “get_train_configs(self) -> 'list[TrainConfig]'”

Expand the grid into validated, ready-to-run TrainConfigs.

iter_variants(self) -> 'list[tuple[dict[str, Any], TrainConfig]]'

Section titled “iter_variants(self) -> 'list[tuple[dict[str, Any], TrainConfig]]'”

Return (overrides, config) for each point in the grid.

launch(self, *, continue_on_error: 'bool' = True, prepare_inputs: 'bool' = False) -> 'list[TrainingRun]'

Section titled “launch(self, *, continue_on_error: 'bool' = True, prepare_inputs: 'bool' = False) -> 'list[TrainingRun]'”

Start every variant as a detached Modal call and return immediately.

train(self, *, max_parallel: 'int' = 1, continue_on_error: 'bool' = True) -> 'list[TrainResult]'

Section titled “train(self, *, max_parallel: 'int' = 1, continue_on_error: 'bool' = True) -> 'list[TrainResult]'”

Run every variant and return the successful TrainResults.

Source: modal_training_gym/common/training_group.py