nat.experimental.test_time_compute.editing.iterative_plan_refinement_editor#

Attributes#

logger

Classes#

IterativePlanRefinementEditor

A planner that generates an initial plan, then refines it multiple times

Functions#

register_iterative_plan_refinement_editor(config, builder)

Module Contents#

logger#

class IterativePlanRefinementEditor( config: nat.data_models.ttc_strategy.TTCStrategyBaseConfig, )#

Bases: nat.experimental.test_time_compute.models.strategy_base.StrategyBase

A planner that generates an initial plan, then refines it multiple times using the same LLM. Each iteration updates the plan to (hopefully) be better.

llm_bound = None#

supported_pipeline_types() → [nat.experimental.test_time_compute.models.stage_enums.PipelineTypeEnum]#: Return the stage types supported by this selector.

stage_type() → nat.experimental.test_time_compute.models.stage_enums.StageTypeEnum#: Return the stage type of this strategy.

async build_components(builder: nat.builder.builder.Builder) → None#: Build the components required for the iterative planner.

async refine_single( prompt: str, context: str, ttc_item: nat.experimental.test_time_compute.models.ttc_item.TTCItem, prompt_idx: int, ) → nat.experimental.test_time_compute.models.ttc_item.TTCItem#

async ainvoke(

items: list[nat.experimental.test_time_compute.models.ttc_item.TTCItem],

original_prompt: str | None = None,

agent_context: str | None = None,

**kwargs,

) → list[nat.experimental.test_time_compute.models.ttc_item.TTCItem]#

Runs the iterative plan refinement process on the provided planning items.

Each planning item is refined in parallel the configured number of times. Default is 3.

Args:: items (list[TTCItem]): The planning items to refine. original_prompt (str): The original prompt used to generate the plans. agent_context (str): The context for the agent.
Returns:: list[TTCItem]: The refined planning items.

async register_iterative_plan_refinement_editor( config: nat.experimental.test_time_compute.models.editor_config.IterativePlanRefinementConfig, builder: nat.builder.builder.Builder, )#

Args:: config (IterativePlanRefinementConfig): The configuration for the strategy.
Returns:: IterativePlanRefinementEditor: The registered strategy instance.