"```python\nfrom modal_training_gym.common.train_result import TrainResult\n```\n\nOne completed training run's checkpoint handle.\n\nConstructed and persisted by each framework's `train` function at\nthe end of a run, and loaded by eval scripts via :meth:`load`. All\nfields are pure data — no method connects to Modal until explicitly\ninvoked.\n\n## Constructor\n\n```python\nTrainResult(app_name, framework, training_run_id, checkpoint_dir='', checkpoints_volume_name='', checkpoints_mount_path='', model_config=None, wandb_project='', wandb_entity='', wandb_training_run_id='', group_id='', extra=<factory>)\n```\n\n| Parameter | Type | Default | Description |\n|-----------|------|---------|-------------|\n| `app_name` | `str` | *required* | |\n| `framework` | `Framework` | *required* | |\n| `training_run_id` | `str` | *required* | |\n| `checkpoint_dir` | `str` | `\"\"` | |\n| `checkpoints_volume_name` | `str` | `\"\"` | |\n| `checkpoints_mount_path` | `str` | `\"\"` | |\n| `model_config` | `'ModelConfig \\| None'` | `None` | |\n| `wandb_project` | `str` | `\"\"` | |\n| `wandb_entity` | `str` | `\"\"` | |\n| `wandb_training_run_id` | `str` | `\"\"` | |\n| `group_id` | `str` | `\"\"` | |\n| `extra` | `dict[str, Any]` | `<factory>` | |\n\n## Attributes\n\n| Attribute | Type | Default | Description |\n|-----------|------|---------|-------------|\n| `app_name` | `str` | | |\n| `framework` | `Framework` | | |\n| `training_run_id` | `str` | | |\n| `checkpoint_dir` | `str` | `\"\"` | |\n| `checkpoints_volume_name` | `str` | `\"\"` | |\n| `checkpoints_mount_path` | `str` | `\"\"` | |\n| `model_config` | `'ModelConfig \\| None'` | `None` | |\n| `wandb_project` | `str` | `\"\"` | |\n| `wandb_entity` | `str` | `\"\"` | |\n| `wandb_training_run_id` | `str` | `\"\"` | |\n| `group_id` | `str` | `\"\"` | |\n| `extra` | `dict[str, Any]` | `{}` | |\n\n## Methods\n\n### `from_training_run_id(training_run_id: 'str') -> \"'TrainResult'\"`\n\nLoad a completed run's result.\n\n### `load(training_run_id: 'str') -> \"'TrainResult'\"`\n\n### `save(self, *, is_async: 'bool' = False) -> 'None | Awaitable[None]'`\n\n### `volume(self) -> \"'Volume'\"`\n\n## Related Tutorials\n\n- [Qwen3-4B haiku evaluation with verifiable rewards — serve, evaluate, train, compare](/tutorials/rl/000_rl_basics/)\n\n**Source:** [`modal_training_gym/common/train_result.py`](https://github.com/modal-projects/training-gym/blob/main/modal_training_gym/common/train_result.py)\n"

TrainResult

One completed training run's checkpoint handle.

from modal_training_gym.common.train_result import TrainResult

One completed training run’s checkpoint handle.

Constructed and persisted by each framework’s train function at the end of a run, and loaded by eval scripts via :meth:load. All fields are pure data — no method connects to Modal until explicitly invoked.

Constructor

TrainResult(app_name, framework, training_run_id, checkpoint_dir='', checkpoints_volume_name='', checkpoints_mount_path='', model_config=None, wandb_project='', wandb_entity='', wandb_training_run_id='', group_id='', extra=<factory>)

Parameter	Type	Default
`app_name`	`str`	required
`framework`	`Framework`	required
`training_run_id`	`str`	required
`checkpoint_dir`	`str`	`""`
`checkpoints_volume_name`	`str`	`""`
`checkpoints_mount_path`	`str`	`""`
`model_config`	`'ModelConfig \| None'`	`None`
`wandb_project`	`str`	`""`
`wandb_entity`	`str`	`""`
`wandb_training_run_id`	`str`	`""`
`group_id`	`str`	`""`
`extra`	`dict[str, Any]`	`<factory>`

Attributes

Attribute	Type	Default
`app_name`	`str`
`framework`	`Framework`
`training_run_id`	`str`
`checkpoint_dir`	`str`	`""`
`checkpoints_volume_name`	`str`	`""`
`checkpoints_mount_path`	`str`	`""`
`model_config`	`'ModelConfig \| None'`	`None`
`wandb_project`	`str`	`""`
`wandb_entity`	`str`	`""`
`wandb_training_run_id`	`str`	`""`
`group_id`	`str`	`""`
`extra`	`dict[str, Any]`	`{}`

Methods

`from_training_run_id(training_run_id: 'str') -> "'TrainResult'"`

Load a completed run’s result.

`load(training_run_id: 'str') -> "'TrainResult'"`

`save(self, *, is_async: 'bool' = False) -> 'None | Awaitable[None]'`

`volume(self) -> "'Volume'"`

Qwen3-4B haiku evaluation with verifiable rewards — serve, evaluate, train, compare

Source: modal_training_gym/common/train_result.py