nemo_gym.tasks.harbor

Read Harbor task folders and hub references into NeMo Gym materialized tasks.

Harbor is the input format; Gym owns execution. A task is a folder holding task.toml, instruction.md, environment/ and tests/. A dataset is a folder of task folders. harbor:<dataset>[@<version>] names a dataset in the Harbor registry, which this package fetches into Gym’s shared datasets folder.

Submodules

Package Contents

Classes

Name	Description
`HarborTask`	A task folder read into memory.
`HarborTaskConfig`	One `task.toml`.

Functions

Name	Description
`content_hash`	Digest the task folder at `path`.
`discover_tasks`	A folder with `task.toml` is one task; otherwise its direct children are the tasks.
`load_task`	Read `path` as one task. Raises `HarborTaskError` when it is not one.

Data

DIGEST_KEY

API

class nemo_gym.tasks.harbor.HarborTask(
    path: pathlib.Path,
    task_id: str,
    config: nemo_gym.tasks.harbor.models.HarborTaskConfig,
    instruction: str,
    digest: str,
    image: str | None,
    workdir: str | None,
    env: dict[str, str],
    user: str | None
)

Dataclass

A task folder read into memory.

task_id is the folder name: [task].name is an org/name label that is not unique across a dataset. image is the prebuilt image when the task declares one, else the base image of a base-image-only Dockerfile, else None (a build is required). Without a prebuilt image, workdir, env and user merge task.toml with what the Dockerfile recorded; task.toml wins. With a prebuilt image, the Dockerfile contributes no settings.

config

HarborTaskConfig

digest

str

env

dict[str, str]

has_solution

bool

image

str | None

instruction

str

needs_sandbox

bool

A task runs in a sandbox exactly when it declares an image or a Dockerfile.

path

Path

task_id

str

user

str | None

workdir

str | None

class nemo_gym.tasks.harbor.HarborTaskConfig()

Bases: HarborSettings

One task.toml.

agent

HarborAgent = Field(default_factory=HarborAgent)

artifacts

list[Any] = Field(default_factory=list)

deprecated_keys

frozenset[str]

environment

HarborEnvironment = Field(default_factory=HarborEnvironment)

is_shared_verifier

bool

Harbor runs the verifier in the agent’s container unless a separate one is declared.

metadata

dict[str, Any] = Field(default_factory=dict)

schema_version

str = '1.4'

solution

HarborSolution = Field(default_factory=HarborSolution)

source

str | None = None

task

dict[str, Any] = Field(default_factory=dict)

verifier

HarborVerifier = Field(default_factory=HarborVerifier)

nemo_gym.tasks.harbor.HarborTaskConfig.read_deprecated_fields(
    data: typing.Any
) -> typing.Any

classmethod

nemo_gym.tasks.harbor.content_hash(
    path: pathlib.Path
) -> str

Digest the task folder at path.

nemo_gym.tasks.harbor.discover_tasks(
    root: pathlib.Path,
    skipped: dict[str, nemo_gym.tasks.harbor.task.HarborTaskError] | None = None
) -> list[nemo_gym.tasks.harbor.task.HarborTask]

A folder with task.toml is one task; otherwise its direct children are the tasks.

With skipped, a child that fails to load is recorded there under its folder name and the others still load; without it the first failure raises. A root that is itself a task always raises.

nemo_gym.tasks.harbor.load_task(
    path: pathlib.Path
) -> nemo_gym.tasks.harbor.task.HarborTask

Read path as one task. Raises HarborTaskError when it is not one.

nemo_gym.tasks.harbor.digest.DIGEST_KEY = '_ng_digest'