nat.experimental.test_time_compute.models.selection_config

Classes

LLMBasedPlanSelectionConfig	Configuration for LLMBasedSelection.
LLMBasedAgentOutputSelectionConfig	Configuration for LLMBasedSelection.
LLMBasedOutputMergingConfig	Configuration for LLMBasedSelection.
ThresholdSelectionConfig	Configuration for a selection strategy that keeps only the items
BestOfNSelectionConfig	Configuration for Best of N Selection
LLMJudgeSelectionConfig	Configuration for a judge-based selection strategy.

Module Contents

class LLMBasedPlanSelectionConfig

Bases: nat.data_models.ttc_strategy.TTCStrategyBaseConfig

Configuration for LLMBasedSelection.

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

selection_template: str = None

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

Ensure that the selection_llm is provided when using LLMBasedSelection.

class LLMBasedAgentOutputSelectionConfig

Bases: nat.data_models.ttc_strategy.TTCStrategyBaseConfig

Configuration for LLMBasedSelection.

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

selection_template: str = None

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

Ensure that the selection_llm is provided when using LLMBasedSelection.

class LLMBasedOutputMergingConfig

Bases: nat.data_models.ttc_strategy.TTCStrategyBaseConfig

Configuration for LLMBasedSelection.

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

selection_template: str = None

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

Ensure that the selection_llm is provided when using LLMBasedSelection.

class ThresholdSelectionConfig

Bases: nat.data_models.ttc_strategy.TTCStrategyBaseConfig

Configuration for a selection strategy that keeps only the items whose scores exceed a specified threshold.

threshold: float = None

class BestOfNSelectionConfig

Bases: nat.data_models.ttc_strategy.TTCStrategyBaseConfig

Configuration for Best of N Selection

class LLMJudgeSelectionConfig

Bases: nat.data_models.ttc_strategy.TTCStrategyBaseConfig

Configuration for a judge-based selection strategy.

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

selection_template: str = None

classmethod validate_config(values: dict) → dict