GraphGen released: LLM training data built from knowledge graphs

GraphGen, an open-source framework that uses knowledge graphs to generate synthetic data for supervised fine-tuning (SFT) of LLMs, has been released. It builds a fine-grained knowledge graph from source text, identifies a model's knowledge gaps, and then generates targeted QA data aimed at filling them. The code is available on GitHub, and a paper describing the method can be read on arXiv.