`nemo_gym.rollout_collection`#

Module Contents#

Classes#

`SharedRolloutCollectionConfig`
`E2ERolloutCollectionConfig`	Spin up all necessary servers and perform a batch of rollout collection using each dataset inside the provided configs.
`RolloutCollectionConfig`	Perform a batch of rollout collection.
`RolloutCollectionHelper`

Functions#

collect_rollouts

API#

class nemo_gym.rollout_collection.SharedRolloutCollectionConfig(/, **data: typing.Any)[source]#

Bases: nemo_gym.config_types.BaseNeMoGymCLIConfig

output_jsonl_fpath: str#: ‘Field(…)’

num_samples_in_parallel: Optional[int]#: ‘Field(…)’

responses_create_params: Dict[str, Any]#: ‘Field(…)’

class nemo_gym.rollout_collection.E2ERolloutCollectionConfig(/, **data: typing.Any)[source]#

Bases: nemo_gym.rollout_collection.SharedRolloutCollectionConfig

Spin up all necessary servers and perform a batch of rollout collection using each dataset inside the provided configs.

Examples:

ng_collect_rollouts         +output_jsonl_fpath=weather_rollouts.jsonl         +num_samples_in_parallel=10

Initialization

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

split: Union[Literal[train], Literal[validation]]#: None

class nemo_gym.rollout_collection.RolloutCollectionConfig(/, **data: typing.Any)[source]#

Bases: nemo_gym.rollout_collection.SharedRolloutCollectionConfig

Perform a batch of rollout collection.

Examples:

ng_collect_rollouts         +agent_name=example_single_tool_call_simple_agent         +input_jsonl_fpath=weather_query.jsonl         +output_jsonl_fpath=weather_rollouts.jsonl         +limit=100         +num_repeats=4         +num_samples_in_parallel=10

Initialization

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

agent_name: Optional[str]#: ‘Field(…)’

input_jsonl_fpath: str#: ‘Field(…)’

limit: Optional[int]#: ‘Field(…)’

num_repeats: Optional[int]#: ‘Field(…)’

num_repeats_add_seed: bool#: ‘Field(…)’

resume_from_cache: bool#: ‘Field(…)’

property materialized_jsonl_fpath: pathlib.Path#

class nemo_gym.rollout_collection.RolloutCollectionHelper(/, **data: typing.Any)[source]#

Bases: pydantic.BaseModel

_preprocess_rows_from_config( config: nemo_gym.rollout_collection.RolloutCollectionConfig, ) → List[Dict][source]#

_load_from_cache( config: nemo_gym.rollout_collection.RolloutCollectionConfig, ) → Tuple[List[Dict], List[Dict], List[Dict], List[List[str]]][source]#

async run_from_config( config: nemo_gym.rollout_collection.RolloutCollectionConfig, ) → Tuple[List[Dict]][source]#

run_examples( examples: List[Dict], head_server_config: Optional[nemo_gym.config_types.BaseServerConfig] = None, semaphore: Optional[asyncio.Semaphore] = None, ) → Iterator[asyncio.Future][source]#: We provide this function as a lower level interface for running rollout collection.

setup_server_client( head_server_config: Optional[nemo_gym.config_types.BaseServerConfig] = None, ) → nemo_gym.server_utils.ServerClient[source]#

nemo_gym.rollout_collection.collect_rollouts()[source]#