nemo_curator.backends.failed_task_markers

View as Markdown

Module Contents

Functions

NameDescription
_configure_failed_task_manifest_dir-
configure_failed_task_manifest_dirConfigure a local attempt-scoped FailedTask manifest directory unless overridden.
configure_slurm_array_failed_task_manifest_dirConfigure an attempt-scoped FailedTask manifest directory unless overridden.
failed_task_manifest_existsReturn whether the current attempt has recorded any FailedTask.
record_failed_tasksWrite one attempt-scoped manifest after any FailedTask is detected.

Data

FAILED_TASKS_DIR_ENV_VAR

FAILED_TASK_MANIFEST_FILENAME

API

nemo_curator.backends.failed_task_markers._configure_failed_task_manifest_dir(
default_dir: pathlib.Path
) -> pathlib.Path
nemo_curator.backends.failed_task_markers.configure_failed_task_manifest_dir(
checkpoint_path: str | pathlib.Path
) -> pathlib.Path

Configure a local attempt-scoped FailedTask manifest directory unless overridden.

nemo_curator.backends.failed_task_markers.configure_slurm_array_failed_task_manifest_dir(
checkpoint_path: str | pathlib.Path,
shard_index: int
) -> pathlib.Path

Configure an attempt-scoped FailedTask manifest directory unless overridden.

nemo_curator.backends.failed_task_markers.failed_task_manifest_exists(
manifest_dir: str | pathlib.Path | None = None
) -> bool

Return whether the current attempt has recorded any FailedTask.

nemo_curator.backends.failed_task_markers.record_failed_tasks() -> None

Write one attempt-scoped manifest after any FailedTask is detected.

nemo_curator.backends.failed_task_markers.FAILED_TASKS_DIR_ENV_VAR = 'NEMO_CURATOR_FAILED_TASKS_DIR'
nemo_curator.backends.failed_task_markers.FAILED_TASK_MANIFEST_FILENAME = 'failed_tasks.json'