TrainConfig

"# TrainConfig\n\n```python\nfrom modal_training_gym.common.train import TrainConfig\n```\n\nA dataset, model, and recipe for one training run.\n\n**Attributes**\n\n<div class=\"tg-param\">\n<p><strong>dataset</strong> <code>DatasetConfig</code></p>\n<p>Training dataset materialized in the framework's <code>/data</code> volume.</p>\n</div>\n\n<div class=\"tg-param\">\n<p><strong>model</strong> <code>ModelConfig</code></p>\n<p>Model identity and weight download behavior.</p>\n</div>\n\n<div class=\"tg-param\">\n<p><strong>recipe</strong> <code>SlimeRecipe | MilesRecipe</code></p>\n<p>Training framework, Modal resources, and framework arguments.</p>\n</div>\n\n<div class=\"tg-param\">\n<p><strong>checkpoint</strong> <code>Checkpoint | None</code></p>\n<p>Megatron checkpoint to resume from. <code>model</code> remains the source for tokenizer and architecture metadata.</p>\n</div>\n\n<div class=\"tg-param\">\n<p><strong>detach</strong> <code>bool</code></p>\n<p>Keep training on Modal if the local <code>train()</code> wait is interrupted. <code>False</code> stops the app. <span class=\"tg-param-default\">Default: True</span></p>\n</div>\n\n<div class=\"tg-param\">\n<p><strong>group_id</strong> <code>str | None</code></p>\n<p>Sweep ID assigned by <code>TrainingGroup</code>.</p>\n</div>\n\n<div class=\"tg-param\">\n<p><strong>group_overrides</strong> <code>dict[str, Any] | None</code></p>\n<p>Variant overrides keyed by dotted field path.</p>\n</div>\n\n<div class=\"tg-param\">\n<p><strong>group_axes</strong> <code>list[str] | None</code></p>\n<p>Swept field paths that group dashboard variants.</p>\n</div>\n\n## `context_plan_line`\n\n```python\ncontext_plan_line() -> str | None\n```\n\nSummarize effective context length and parallelism on one line.\n\n## `framework`\n\n```python\nframework: Framework\n```\n\n## `launch`\n\n```python\nlaunch(*, show_output: bool = True, prepare_inputs: bool = False) -> TrainingRun\n```\n\nStart one training run in a detached Modal app.\n\n**Returns**\n\nThe launched training run.\n\n## `train`\n\n```python\ntrain(*, show_output: bool = True) -> TrainResult\n```\n\nRun one training configuration.\n\n**Returns**\n\nThe completed training result.\n"

from modal_training_gym.common.train import TrainConfig

A dataset, model, and recipe for one training run.

Attributes

dataset DatasetConfig

Training dataset materialized in the framework's /data volume.

model ModelConfig

Model identity and weight download behavior.

recipe SlimeRecipe | MilesRecipe

Training framework, Modal resources, and framework arguments.

checkpoint Checkpoint | None

Megatron checkpoint to resume from. model remains the source for tokenizer and architecture metadata.

detach bool

Keep training on Modal if the local train() wait is interrupted. False stops the app. Default: True

group_id str | None

Sweep ID assigned by TrainingGroup.

group_overrides dict[str, Any] | None

Variant overrides keyed by dotted field path.

group_axes list[str] | None

Swept field paths that group dashboard variants.

`context_plan_line`

context_plan_line() -> str | None

Summarize effective context length and parallelism on one line.

`framework`

framework: Framework

`launch`

launch(*, show_output: bool = True, prepare_inputs: bool = False) -> TrainingRun

Start one training run in a detached Modal app.

Returns

The launched training run.

`train`

train(*, show_output: bool = True) -> TrainResult

Run one training configuration.

Returns

The completed training result.