"```python\nfrom modal_training_gym.common.dataset import HarborDataset\n```\n\nDataset backed by a Harbor task directory structure.\n\nEach task folder contains an instruction file and optional label metadata.\nTasks are discovered by globbing the task_root directory.\n\n**Inherits from:** `DatasetConfig`\n\n## Fields\n\n| Field | Type | Default | Description |\n|-------|------|---------|-------------|\n| `dataset_id` | `str` | `\"\"` | |\n| `input_key` | `str` | `\"\"` | |\n| `label_key` | `str` | `\"\"` | |\n| `apply_chat_template` | `bool` | `True` | |\n| `always_prepare` | `bool` | `False` | |\n| `writes_eval_paths` | `bool` | `True` | |\n| `dataset_name` | `str` | `\"\"` | |\n| `path` | `str \\| None` | `None` | |\n| `task_root` | `str` | `\"\"` | |\n| `task_glob` | `str` | `\"*\"` | |\n| `task_names` | `list[str] \\| None` | `None` | |\n| `instruction_path` | `str` | `\"instruction.md\"` | |\n| `label_metadata_path` | `str \\| None` | `None` | |\n| `test_data_dir` | `str \\| None` | `None` | |\n| `output_format` | `str` | `\"parquet\"` | |\n| `prompt_template` | `str` | `\"{instruction}\"` | |\n| `system_prompt` | `str` | `\"\"` | |\n| `train_size` | `int \\| None` | `None` | |\n| `eval_size` | `int \\| None` | `None` | |\n| `train_repeats` | `int` | `1` | |\n| `eval_repeats` | `int` | `1` | |\n| `shuffle_tasks` | `bool` | `False` | |\n| `shuffle_seed` | `int` | `0` | |\n\n## Methods\n\n### `load(self, split: \"Literal['all', 'train', 'eval']\" = 'all') -> 'Any'`\n\nLoad raw examples, optionally filtered by split.\n\n### `prepare(self, path: 'str', eval_paths: 'dict[str, str] | None' = None) -> 'None'`\n\nMaterialize training data to `path` (and eval splits to `eval_paths`).\n\n### `to_pandas(self, *, formatted: 'bool' = False)`\n\n### `validate_prepared(self, path: 'str') -> 'None'`\n\nSniff what `prepare()` wrote and confirm the columns the framework will index.\n\n## Related Tutorials\n\n- [Code RL with Harbor hello-world and sandboxed verification](/tutorials/rl/001_sandboxes/)\n\n**Source:** [`modal_training_gym/common/dataset.py`](https://github.com/modal-projects/training-gym/blob/main/modal_training_gym/common/dataset.py)\n"

HarborDataset

Dataset backed by a Harbor task directory structure.

from modal_training_gym.common.dataset import HarborDataset

Dataset backed by a Harbor task directory structure.

Each task folder contains an instruction file and optional label metadata. Tasks are discovered by globbing the task_root directory.

Inherits from: DatasetConfig

Fields

Field	Type	Default
`dataset_id`	`str`	`""`
`input_key`	`str`	`""`
`label_key`	`str`	`""`
`apply_chat_template`	`bool`	`True`
`always_prepare`	`bool`	`False`
`writes_eval_paths`	`bool`	`True`
`dataset_name`	`str`	`""`
`path`	`str \| None`	`None`
`task_root`	`str`	`""`
`task_glob`	`str`	`"*"`
`task_names`	`list[str] \| None`	`None`
`instruction_path`	`str`	`"instruction.md"`
`label_metadata_path`	`str \| None`	`None`
`test_data_dir`	`str \| None`	`None`
`output_format`	`str`	`"parquet"`
`prompt_template`	`str`	`"{instruction}"`
`system_prompt`	`str`	`""`
`train_size`	`int \| None`	`None`
`eval_size`	`int \| None`	`None`
`train_repeats`	`int`	`1`
`eval_repeats`	`int`	`1`
`shuffle_tasks`	`bool`	`False`
`shuffle_seed`	`int`	`0`

Methods

`load(self, split: "Literal['all', 'train', 'eval']" = 'all') -> 'Any'`

Load raw examples, optionally filtered by split.

`prepare(self, path: 'str', eval_paths: 'dict[str, str] | None' = None) -> 'None'`

Materialize training data to path (and eval splits to eval_paths).

`to_pandas(self, *, formatted: 'bool' = False)`

`validate_prepared(self, path: 'str') -> 'None'`

Sniff what prepare() wrote and confirm the columns the framework will index.

Code RL with Harbor hello-world and sandboxed verification

Source: modal_training_gym/common/dataset.py