TrainingGroup
"# TrainingGroup\n\n```python\nfrom modal_training_gym.common.training_group import TrainingGroup\n```\n\nA parameter sweep over a base `TrainConfig`.\n\n## Constructor\n\n```python\nTrainingGroup(base: TrainConfig, grid: dict[str, list[Any]] | None = None, *, name: str | None = None) -> None\n```\n\n## `get_train_configs`\n\n```python\nget_train_configs() -> list[TrainConfig]\n```\n\nBuild validated configs for each sweep point.\n\n**Returns**\n\nValidated training configs.\n\n## `iter_variants`\n\n```python\niter_variants() -> list[tuple[dict[str, Any], TrainConfig]]\n```\n\nExpand the sweep grid.\n\n**Returns**\n\nPairs of overrides and validated training configs.\n\n## `launch`\n\n```python\nlaunch(*, continue_on_error: bool = True, prepare_inputs: bool = False) -> list[TrainingRun]\n```\n\nLaunch every variant as a detached Modal call.\n\n**Parameters**\n\n<div class=\"tg-param\">\n<p><strong>continue_on_error</strong> <code>bool</code></p>\n<p>Continue after a variant fails to launch. <span class=\"tg-param-default\">Default: True</span></p>\n</div>\n\n<div class=\"tg-param\">\n<p><strong>prepare_inputs</strong> <code>bool</code></p>\n<p>Materialize model and dataset inputs before launching. <span class=\"tg-param-default\">Default: False</span></p>\n</div>\n\n**Returns**\n\nLaunched training runs.\n\n## `train`\n\n```python\ntrain(*, max_parallel: int = 1, continue_on_error: bool = True) -> list[TrainResult]\n```\n\nTrain every variant.\n\n**Parameters**\n\n<div class=\"tg-param\">\n<p><strong>max_parallel</strong> <code>int</code></p>\n<p>Maximum number of variants to run at once. <span class=\"tg-param-default\">Default: 1</span></p>\n</div>\n\n<div class=\"tg-param\">\n<p><strong>continue_on_error</strong> <code>bool</code></p>\n<p>Continue after a variant fails. <span class=\"tg-param-default\">Default: True</span></p>\n</div>\n\n**Returns**\n\nSuccessful training results.\n"
from modal_training_gym.common.training_group import TrainingGroupA parameter sweep over a base TrainConfig.
Constructor
Section titled “Constructor”TrainingGroup(base: TrainConfig, grid: dict[str, list[Any]] | None = None, *, name: str | None = None) -> Noneget_train_configs
Section titled “get_train_configs”get_train_configs() -> list[TrainConfig]Build validated configs for each sweep point.
Returns
Validated training configs.
iter_variants
Section titled “iter_variants”iter_variants() -> list[tuple[dict[str, Any], TrainConfig]]Expand the sweep grid.
Returns
Pairs of overrides and validated training configs.
launch
Section titled “launch”launch(*, continue_on_error: bool = True, prepare_inputs: bool = False) -> list[TrainingRun]Launch every variant as a detached Modal call.
Parameters
continue_on_error bool
Continue after a variant fails to launch. Default: True
prepare_inputs bool
Materialize model and dataset inputs before launching. Default: False
Returns
Launched training runs.
train(*, max_parallel: int = 1, continue_on_error: bool = True) -> list[TrainResult]Train every variant.
Parameters
max_parallel int
Maximum number of variants to run at once. Default: 1
continue_on_error bool
Continue after a variant fails. Default: True
Returns
Successful training results.