nemo_automodel.components.launcher.nemo_run.config#

Module Contents#

Classes#

NemoRunConfig

Configuration for the NeMo-Run launcher backend.

Data#

API#

nemo_automodel.components.launcher.nemo_run.config._NEMORUN_HOME#

‘get(…)’

nemo_automodel.components.launcher.nemo_run.config.DEFAULT_EXECUTORS_FILE#

‘join(…)’

nemo_automodel.components.launcher.nemo_run.config._LAUNCHER_KEYS#

‘frozenset(…)’

class nemo_automodel.components.launcher.nemo_run.config.NemoRunConfig[source]#

Configuration for the NeMo-Run launcher backend.

The executor field selects a named executor from $NEMORUN_HOME/executors.py, or "local" for local execution.

Any key not recognised as a launcher setting is collected into overrides and applied directly to the executor via setattr. This means any executor attribute (nodes, partition, container_image, time, env_vars, etc.) can be overridden from YAML without changes to this config class.

executor: str#

‘local’

job_name: str = <Multiline-String>#
detach: bool#

True

tail_logs: bool#

False

executors_file: str#

‘field(…)’

job_dir: str = <Multiline-String>#
overrides: dict#

‘field(…)’

classmethod from_dict(
d: dict,
) nemo_automodel.components.launcher.nemo_run.config.NemoRunConfig[source]#

Build from a raw YAML dict, splitting launcher keys from executor overrides.