nat.experimental.test_time_compute.models.scoring_config

Classes

LLMBasedPlanScoringConfig	Configuration for LLMBasedScoring.
LLMBasedAgentScoringConfig	Configuration for LLMBasedScoring.
MotivationAwareScoringConfig	Configuration for a scoring strategy that considers both the original input (task)

Module Contents

class LLMBasedPlanScoringConfig

Bases: nat.data_models.ttc_strategy.TTCStrategyBaseConfig

Configuration for LLMBasedScoring.

scoring_llm: nat.data_models.component_ref.LLMRef | Any | None = None

scoring_template: str = None

validate_strategies(values: dict[str, Any]) → dict[str, Any]

Ensure that the scoring_llm is provided when using LLMBasedScoring.

class LLMBasedAgentScoringConfig

Bases: nat.data_models.ttc_strategy.TTCStrategyBaseConfig

Configuration for LLMBasedScoring.

scoring_llm: nat.data_models.component_ref.LLMRef | Any | None = None

scoring_template: str = None

validate_strategies(values: dict[str, Any]) → dict[str, Any]

Ensure that the scoring_llm is provided when using LLMBasedScoring.

class MotivationAwareScoringConfig

Bases: nat.data_models.ttc_strategy.TTCStrategyBaseConfig

Configuration for a scoring strategy that considers both the original input (task) and the motivation (from metadata) along with the current output.

scoring_llm: nat.data_models.component_ref.LLMRef | None = None

scoring_template: str = None

validate_scoring_llm(values)