GraphGen

KG-guided synthetic SFT data: builds a knowledge graph from source text, finds the LLM's knowledge gaps via calibration error, and generates targeted long-tail QA pairs. ACL-published.

1,159 85 Python Apache-2.0updated 2 months ago

training

View on GitHubDispute this mapping →

Curator's take

The interesting bit isn't "synthetic QA" — it's targeting: GraphGen measures the student model's expected calibration error to find what it *doesn't* know, then samples multi-hop KG neighborhoods to generate exactly those long-tail QA pairs. Use it when domain-adapting a model and naive synthetic data keeps rehearsing what the model already knows. Output feeds straight into LLaMA-Factory or xtuner. NOT a general data-prep system — it makes QA-style SFT/pretrain data from documents, nothing else; for cleaning, filtering and pipeline plumbing you still want an operator system like DataFlow.

Mapped by ShipWithAI editors · links verified

README.md

GraphGen: Enhancing Supervised Fine-Tuning for LLMs with Knowledge-Driven Synthetic Data Generation

English | 中文

📚 Table of Contents

📝 What is GraphGen?
📌 Latest Updates
⚙️ Support List
🚀 Quick Start
🏗️ System Architecture
🍀 Acknowledgements
📚 Citation
📜 License
📅 Star History

📝 What is GraphGen?

GraphGen is a framework for synthetic data generation guided by knowledge graphs. Please check the paper and best practice.

It begins by constructing a fine-grained knowledge graph from the source text，then identifies knowledge gaps in LLMs using the expected calibration error metric, prioritizing the generation of QA pairs that target high-value, long-tail knowledge. Furthermore, GraphGen incorporates multi-hop neighborhood sampling to capture complex relational information and employs style-controlled generation to diversify the resulting QA data.

After data generation, you can use LLaMA-Factory and xtuner to finetune your LLMs.

📌 Latest Updates

🎉 2026.04.13: The paper based on GraphGen, Knowledge-to-Verification: Exploring RLVR for LLMs in Knowledge-Intensive Domains, has been accepted to the ACL 2026 Main Conference! Congratulations! [arXiv][Code]
2026.02.04: We support HuggingFace Datasets as input data source for data generation now.
2026.01.15: LLM benchmark synthesis now supports single/multiple-choice & fill-in-the-blank & true-or-false—ideal for education 🌟🌟

History

2025.12.26: Knowledge graph evaluation metrics about accuracy (entity/relation), consistency (conflict detection), structural robustness (noise, connectivity, degree distribution)
2025.12.16: Added rocksdb for key-value storage backend and kuzudb for g

Continue your stack

What teams reach for next — and why each earns a place beside GraphGen. Ranked by curator confidence.

Pairs well with

LlamaFactory

The unified fine-tuning framework: 100+ LLMs and VLMs via LoRA/QLoRA/full-parameter, config-driven or through the LlamaBoard GUI. ACL 2024, 1000+ citations, 73k stars.

Why it fitsGraphGen's README names LLaMA-Factory as the intended next step: generate gap-targeted QA data with GraphGen, then fine-tune any of LLaMA-Factory's 100+ models on it via LoRA/QLoRA.

trl

Hugging Face's post-training library: SFT, DPO, GRPO, KTO and reward-model trainers on top of Transformers — from a Colab LoRA run to multi-GPU deployments.

Why it fitsGraphGen sits on the data side of post-training; its QA output is standard SFT material for TRL's SFTTrainer if you'd rather stay in the Hugging Face stack than use LLaMA-Factory.

Alternative to See all + compare →

DataFlow

Operator-based system for LLM data prep — 100+ operators composed into pipelines that generate, clean, evaluate and filter pretraining/SFT/RL data, with a WebUI and a pipeline-building agent.

Why it fitsBoth produce SFT training data, differently: DataFlow is a general operator-pipeline system for generating/cleaning/filtering at scale; GraphGen is one focused method — KG + calibration-error targeting of long-tail knowledge. DataFlow to industrialize a data pipeline, GraphGen when gap coverage is the point.