Implementation:Ucbepic Docetl MapOperation Execute

Knowledge Sources	DocETL
Domains	NLP, LLM_Operations
Last Updated	2026-02-08 01:40 GMT

Overview

Concrete operation for applying LLM transformations to individual documents or chunks provided by DocETL's operations module.

Description

MapOperation processes each input document independently through an LLM using a Jinja2 prompt template and structured output schema. It supports gleaning (iterative validation), batching (multiple documents per call), parallel execution, calibration, and tool use. The operation is the most widely used in DocETL pipelines.

Usage

Use MapOperation for per-document or per-chunk LLM processing. In a chunking pipeline, it processes enriched chunks after GatherOperation. It can also be used standalone for simple document transformations.

Code Reference

Source Location

Repository: docetl
File: docetl/operations/map.py
Lines: L23-857

Signature

class MapOperation(BaseOperation):
    class schema(BaseOperation.schema):
        type: str = "map"
        output: dict[str, Any] | None = None
        prompt: str | None = None
        model: str | None = None
        optimize: bool | None = None
        batch_size: int | None = None
        gleaning: dict | None = None
        timeout: int | None = None
        litellm_completion_kwargs: dict[str, Any] = {}

    def execute(self, input_data: list[dict]) -> tuple[list[dict], float]:
        """Process each document via LLM. Returns (results, total_cost)."""

Import

from docetl.operations.map import MapOperation

I/O Contract

Inputs

Name	Type	Required	Description
prompt	str	Yes	Jinja2 template with Template:Input.field variables
output.schema	dict	Yes	Expected output field names and types
model	str	No	LLM model name (defaults to pipeline default)
input_data	list[dict]	Yes	Documents or chunks to process

Outputs

Name	Type	Description
results	list[dict]	Documents with LLM-generated fields added
cost	float	Total LLM API cost

Usage Examples

operations:
  - name: extract_info
    type: map
    prompt: |
      Extract key information from this text chunk:
      {{ input.content_chunk_rendered }}
    output:
      schema:
        key_findings: "list[str]"
        entities: "list[str]"
    model: gpt-4o

Related Pages

Implements Principle

Principle:Ucbepic_Docetl_Chunk_Processing

Requires Environment

Uses Heuristic

Page Connections

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

Principle

Implementation

Heuristic

Environment