synthetic-data

What problem does it solve?

Generating diverse, labeled synthetic test inputs is essential to stress-test eval pipelines, expand coverage, and ensure robust benchmarking across ML models.

Core Features & Use Cases

• Defines multiple variation dimensions (topic, difficulty, format, length, edge_case) and computes their combinatorial product to maximize coverage.
• Applies filters, sampling, and seeds to guarantee reproducibility and targeted data generation for evaluation and training augmentation.
• Outputs labeled datasets and metadata suitable for eval dashboards, benchmarks, and regression tests.

Quick Start

Invoke /synthetic-data with a dimensions.yaml or inline --dim definitions to generate a batch of synthetic test inputs.