docling-metrics-core 0.14.0

Metric types and evaluation utilities for Docling document processing

docling-metrics-core

A minimal interface for computing and aggregating metrics on paired data samples.

Overview

This package provides base types for building metrics that:

• Evaluate pairs of input samples (e.g., ground-truth vs. prediction)
• Produce per-sample results traceable by sample ID
• Aggregate results across multiple samples

Core Types

• BaseInputSample — Input data with a unique id shared between sample pairs
• BaseSampleResult — Output of evaluating a single sample pair
• BaseAggregateResult — Output of aggregating multiple sample results
• BaseMetric — Abstract interface defining evaluate_sample, aggregate, and evaluate_dataset

Installation

pip install docling-metrics-core

Requirements

• Python >=3.10,<4.0

Usage

Coming soon

Development

# Install in development mode
uv sync

# Run tests
uv run pytest

License

MIT