Implementation:Ucbepic Docetl DSLRunner Load Run Save For Baseline

Knowledge Sources	DocETL DocETL Docs
Domains	Data_Engineering, Optimization
Last Updated	2026-02-08 01:40 GMT

Overview

Concrete tool for running a baseline DocETL pipeline to establish initial metrics before optimization.

Description

This uses the same DSLRunner.load_run_save() method as standard pipeline execution, but in the context of optimization: the pipeline contains operations marked with optimize: True. The baseline run produces initial accuracy and cost metrics that the optimizer uses as a reference point.

Usage

Run the baseline pipeline before invoking the optimizer. The results serve as the "root node" in the MOAR MCTS search tree and the comparison point for V1 optimization candidates.

Code Reference

Source Location

• Repository: docetl
• File: docetl/runner.py
• Lines: L106-130 (__init__), L462-494 (load_run_save)

Signature

class DSLRunner:
    def __init__(self, config: dict, max_threads: int | None = None, **kwargs):
        """Initialize with pipeline config containing optimize: True operations."""

    def load_run_save(self) -> float:
        """Execute baseline pipeline, returning total LLM API cost."""

Import

from docetl.runner import DSLRunner

I/O Contract

Inputs

Name	Type	Required	Description
config	dict	Yes	YAML config with operations marked optimize: True
max_threads	int or None	No	Parallel execution limit

Outputs

Name	Type	Description
load_run_save() returns	float	Total baseline execution cost
output file	JSON or CSV	Baseline results for accuracy measurement

Usage Examples

import yaml
from docetl.runner import DSLRunner

# Load baseline pipeline with optimize flags
with open("baseline_pipeline.yaml") as f:
    config = yaml.safe_load(f)

runner = DSLRunner(config)
baseline_cost = runner.load_run_save()
print(f"Baseline cost: ${baseline_cost:.2f}")

Related Pages

Implements Principle

• Principle:Ucbepic_Docetl_Baseline_Pipeline_Authoring