nemoguardrails.actions.v2_x.generation | NVIDIA NeMo Guardrails Library Developer Guide

A set of actions for generating various types of completions using an LLMs.

Module Contents

Classes

Name	Description
`LLMGenerationActionsV2dotx`	Adapted version of LLMGenerationActions for Colang 2.x.

Functions

Name	Description
`_remove_leading_empty_lines`	Remove the leading empty lines if they exist.

Data

log

API

class nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx()

Bases: LLMGenerationActions

Adapted version of LLMGenerationActions for Colang 2.x.

It overrides some methods.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx._collect_user_intent_and_examples(
    state: nemoguardrails.colang.v2_x.runtime.statemachine.State,
    user_action: str,
    max_example_flows: int
) -> typing.Tuple[typing.List[str], str, bool]

async

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx._ensure_flows_index()

async

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx._ensure_instruction_flows_index()

async

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx._init_colang_flows_index(
    flows: typing.List[str]
) -> typing.Optional[nemoguardrails.embeddings.index.EmbeddingsIndex]

async

Initialize an index with colang flows.

The flows are expected to have full definition.

Args flows: The list of flows, i.e. the flow definition from the source code.

Returns An initialized index.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx._init_flows_index() -> None

async

Initializes the index of flows.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.check_for_active_flow_finished_match(
    state: nemoguardrails.colang.v2_x.runtime.statemachine.State,
    event_name: str,
    arguments: typing.Any = {}
) -> bool

async

Return True if there is a flow waiting for the provided event name and parameters.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.check_if_flow_defined(
    state: nemoguardrails.colang.v2_x.runtime.statemachine.State,
    flow_id: str
) -> bool

async

Return True if a flow with the provided flow_id is defined.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.check_if_flow_exists(
    state: nemoguardrails.colang.v2_x.runtime.statemachine.State,
    flow_id: str
) -> bool

async

Return True if a flow with the provided flow_id exists.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.create_flow(
    events: typing.List[dict],
    name: str,
    body: str,
    decorators: typing.Optional[str] = None
) -> dict

async

Create a new flow during runtime.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.generate_flow(
    state: nemoguardrails.colang.v2_x.runtime.statemachine.State,
    events: typing.List[dict],
    llm: typing.Optional[nemoguardrails.types.LLMModel] = None,
    flow_id: typing.Optional[str] = None
) -> dict

async

Generate the body for a flow.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.generate_flow_continuation(
    state: nemoguardrails.colang.v2_x.runtime.statemachine.State,
    events: typing.List[dict],
    temperature: typing.Optional[float] = None,
    llm: typing.Optional[nemoguardrails.types.LLMModel] = None
) -> dict

async

Generate a continuation for the flow representing the current conversation.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.generate_flow_from_instructions(
    state: nemoguardrails.colang.v2_x.runtime.statemachine.State,
    instructions: str,
    events: typing.List[dict],
    llm: typing.Optional[nemoguardrails.types.LLMModel] = None
) -> dict

async

Generate a flow from the provided instructions.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.generate_flow_from_name(
    state: nemoguardrails.colang.v2_x.runtime.statemachine.State,
    name: str,
    events: typing.List[dict],
    llm: typing.Optional[nemoguardrails.types.LLMModel] = None
) -> str

async

Generate a flow from the provided NAME.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.generate_user_intent(
    state: nemoguardrails.colang.v2_x.runtime.statemachine.State,
    events: typing.List[dict],
    user_action: str,
    max_example_flows: int = 5,
    llm: typing.Optional[nemoguardrails.types.LLMModel] = None
) -> str

async

Generate the canonical form for what the user said i.e. user intent.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.generate_user_intent_and_bot_action(
    state: nemoguardrails.colang.v2_x.runtime.statemachine.State,
    events: typing.List[dict],
    user_action: str,
    max_example_flows: int = 5,
    llm: typing.Optional[nemoguardrails.types.LLMModel] = None
) -> dict

async

Generate the canonical form for what the user said i.e. user intent and a suitable bot action.

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.generate_value(
    state: nemoguardrails.colang.v2_x.runtime.statemachine.State,
    instructions: str,
    events: typing.List[dict],
    var_name: typing.Optional[str] = None,
    llm: typing.Optional[nemoguardrails.types.LLMModel] = None
) -> typing.Any

async

Generate a value in the context of the conversation.

:param instructions: The instructions to generate the value. :param events: The full stream of events so far. :param var_name: The name of the variable to generate. :param llm: Custom llm model to generate_value

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.get_last_user_message(
    events: typing.List[dict],
    llm: typing.Optional[nemoguardrails.types.LLMModel] = None
) -> str

async

nemoguardrails.actions.v2_x.generation.LLMGenerationActionsV2dotx.passthrough_llm_action(
    user_message: str,
    state: nemoguardrails.colang.v2_x.runtime.statemachine.State,
    events: typing.List[dict],
    llm: typing.Optional[nemoguardrails.types.LLMModel] = None
)

async

nemoguardrails.actions.v2_x.generation._remove_leading_empty_lines(
    s: str
) -> str

Remove the leading empty lines if they exist.

A line is considered empty if it has only white spaces.

nemoguardrails.actions.v2_x.generation.log = logging.getLogger(__name__)