Implementation:Ucbepic Docetl MapOptimizer PromptGenerators

Knowledge Sources	Ucbepic_Docetl DocETL Docs
Domains	Data_Processing, Optimization
Last Updated	2026-02-08 00:00 GMT

Overview

Concrete tool for generating validator, combiner, header extraction, schema transform, and improvement prompts for map operation optimization provided by DocETL.

Description

The PromptGenerator class is a component of the MapOptimizer subsystem responsible for synthesizing various prompts needed during the optimization process. It generates validator prompts that assess output quality and completeness, header extraction prompts that identify document structure for chunk processing, combine prompts that merge results from multiple chunks in map-reduce operations, schema transformation prompts that convert parallel map outputs into the target schema, missing-keys synthesis prompts for chain plans with incomplete coverage, and improved operation prompts based on assessment feedback. It also determines whether combine operations are associative (order-independent).

Usage

Use PromptGenerator when you need to create specialized prompts for the map optimization workflow. It is used internally by the MapOptimizer and PlanGenerator to generate validation criteria, chunk combination logic, header detection, schema transformation, and prompt improvement. It is relevant whenever the optimizer needs to assess output quality, combine chunked results, or transform intermediate outputs to match a target schema.

Code Reference

Source Location

Repository: Ucbepic_Docetl
File: docetl/optimizers/map_optimizer/prompt_generators.py
Lines: 1-698

Signature

class PromptGenerator:
    def __init__(
        self,
        runner,
        llm_client: LLMClient,
        console: Console,
        config: dict[str, Any],
        max_threads: int,
        is_filter: bool = False,
    ) -> None: ...

    def _generate_validator_prompt(
        self,
        op_config: dict[str, Any],
        input_data: list[dict[str, Any]],
        output_data: list[dict[str, Any]],
    ) -> str: ...

    def _get_header_extraction_prompt(
        self,
        op_config: dict[str, Any],
        input_data: list[dict[str, Any]],
        split_key: str,
    ) -> tuple[str, dict[str, Any]]: ...

    def _get_improved_prompt(
        self,
        op_config: dict[str, Any],
        assessment: dict[str, Any],
        input_data_sample: list[dict[str, Any]],
    ) -> list[dict[str, Any]]: ...

    def _get_combine_prompt(
        self,
        op_config: dict[str, Any],
        sample_output: list[dict[str, Any]],
    ) -> tuple[str, bool]: ...

    def _edit_subprompt_to_reflect_metadata(
        self,
        subprompt: str,
        metadata_schema: dict[str, Any],
        sample_output: list[dict[str, Any]],
    ) -> str: ...

    def _get_schema_transform_prompt(
        self,
        op_config: dict[str, Any],
        parallel_output_schema: dict[str, Any],
        target_schema: dict[str, Any],
        sample_output: list[dict[str, Any]],
    ) -> str: ...

    def _get_missing_keys_prompt(
        self,
        op_config: dict[str, Any],
        missing_keys: set[str],
        existing_keys: set[str],
    ) -> str: ...

Import

from docetl.optimizers.map_optimizer.prompt_generators import PromptGenerator

I/O Contract

Inputs

Name	Type	Required	Description
runner	Runner	Yes	The pipeline runner object
llm_client	LLMClient	Yes	Client for interacting with the language model
console	Console	Yes	Rich console object for output logging
config	dict[str, Any]	Yes	The pipeline configuration dictionary
max_threads	int	Yes	Maximum number of threads for parallel execution
is_filter	bool	No	Whether the operation is a filter operation (default: False)
op_config	dict[str, Any]	Yes	The operation configuration
input_data	list[dict[str, Any]]	Yes	Input data samples
output_data	list[dict[str, Any]]	Yes	Output data samples for validator prompt generation
split_key	str	Yes	Document split key for header extraction
sample_output	list[dict[str, Any]]	Yes	Sample outputs for combine prompt generation

Outputs

Name	Type	Description
validator_prompt	str	Custom prompt for assessing output quality
header_extraction_prompt	str	Jinja2 template prompt for extracting headers from document chunks
header_output_schema	dict[str, Any]	Output schema for extracted headers
combine_prompt	str	Jinja2 template for combining chunk results in map-reduce
is_associative	bool	Whether the combine operation is order-independent
transform_prompt	str	Prompt for transforming parallel output schema to target schema
synthesis_prompt	str	Prompt for generating missing output keys in chain plans
improved_ops	list[dict[str, Any]]	Operation configs with improved prompts

Usage Examples

from docetl.optimizers.map_optimizer.prompt_generators import PromptGenerator

prompt_gen = PromptGenerator(
    runner=pipeline_runner,
    llm_client=llm_client,
    console=console,
    config=pipeline_config,
    max_threads=4,
)

# Generate a validator prompt
validator_prompt = prompt_gen._generate_validator_prompt(
    op_config=map_op_config,
    input_data=input_samples,
    output_data=output_samples,
)
print(f"Validator prompt: {validator_prompt}")

# Generate a combine prompt for chunk results
combine_prompt, is_associative = prompt_gen._get_combine_prompt(
    op_config=map_op_config,
    sample_output=chunk_outputs,
)
print(f"Combine is associative: {is_associative}")

# Generate header extraction prompt
header_prompt, header_schema = prompt_gen._get_header_extraction_prompt(
    op_config=map_op_config,
    input_data=input_samples,
    split_key="document_text",
)

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment