Implementation:Ucbepic Docetl Build CLI Command

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

Overview

Concrete CLI command and helper for configuring and launching DocETL pipeline optimization.

Description

The docetl build CLI command parses the optimizer_config section from a YAML pipeline file, validates required fields (evaluation_file, metric_key, available_models, max_iterations, save_dir), infers dataset information, and dispatches to either the V1 or MOAR optimizer. The run_moar_optimization() helper function handles MOAR-specific setup.

Usage

Use docetl build pipeline.yaml --optimizer moar to launch MOAR optimization. The pipeline YAML must include an optimizer_config section with all required fields.

Code Reference

Source Location

• Repository: docetl
• File: docetl/cli.py (L19-198), docetl/moar/cli_helpers.py (L92-305)

Signature

# CLI command
def build(
    yaml_file: Path,
    optimizer: str = "moar",
    max_threads: int | None = None,
    resume: bool = False,
    save_path: Path = None,
) -> None:
    """Build/optimize a DocETL pipeline."""

# MOAR helper
def run_moar_optimization(
    yaml_path: str,
    optimizer_config: dict,
) -> Dict[str, Any]:
    """Run MOAR optimization from CLI. Returns experiment summary."""

Import

# CLI usage
docetl build pipeline.yaml --optimizer moar

I/O Contract

Inputs

Name	Type	Required	Description
yaml_file	Path	Yes	Path to YAML pipeline with optimizer_config
optimizer	str	No	"moar" (default) or "v1"
optimizer_config.evaluation_file	str	Yes	Path to @register_eval Python file
optimizer_config.metric_key	str	Yes	Key in evaluation results dict to optimize
optimizer_config.available_models	list[str]	Yes	LLM models to search over
optimizer_config.max_iterations	int	Yes	MCTS iteration budget
optimizer_config.save_dir	str	Yes	Output directory for optimized pipelines

Outputs

Name	Type	Description
results	Dict[str, Any]	Experiment summary with paths to optimized pipelines
optimized YAMLs	files	Written to save_dir

Usage Examples

# Run MOAR optimization
docetl build pipeline.yaml --optimizer moar

# Resume interrupted optimization
docetl build pipeline.yaml --optimizer moar --resume

# optimizer_config section in pipeline.yaml
optimizer_config:
  type: moar
  save_dir: ./moar_results
  available_models:
    - gpt-4o
    - gpt-4o-mini
  evaluation_file: evaluate.py
  metric_key: accuracy
  max_iterations: 40
  rewrite_agent_model: gpt-4o

Related Pages

Implements Principle

• Principle:Ucbepic_Docetl_Optimizer_Configuration