InstructLab Taxonomy

A structured schema for curating the human-authored skills and knowledge used to fine-tune large language models via the LAB methodology.

The InstructLab Taxonomy provides the essential YAML-based framework for the Large-scale Alignment Baseline (LAB) method. It allows contributors to submit diverse training data (spanning compositional skills like writing code to foundational knowledge like world history) without requiring deep machine learning expertise. By organizing data into a clear tree structure of tasks and attributes, the taxonomy enables the synthetic data generation pipeline to produce high-quality instruction-tuning sets. This collaborative approach, backed by IBM and Red Hat, democratizes model alignment by turning community-driven pull requests into measurable improvements for open-source LLMs.

https://github.com/instructlab/taxonomy