graphg 20250416

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

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

📚 Table of Contents

• 📝 What is GraphGen?
• 🚀 Quick Start
• 📌 Latest Updates
• 🌟 Key Features
• 🏗️ System Architecture
• ⚙️ Configurations
• 📅 Roadmap
• 💰 Cost Analysis

📝 What is GraphGen?

GraphGen is a framework for synthetic data generation guided by knowledge graphs. Here is our paper.

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.

🚀 Quick Start

Experience it on the OpenXLab Application Center

Gradio Demo

python webui/app.py

Run from PyPI

1. Install GraphGen

   pip install graphg
   

2. Run in CLI

   SYNTHESIZER_MODEL=your_synthesizer_model_name \
   SYNTHESIZER_BASE_URL=your_base_url_for_synthesizer_model \
   SYNTHESIZER_API_KEY=your_api_key_for_synthesizer_model \
   TRAINEE_MODEL=your_trainee_model_name \
   TRAINEE_BASE_URL=your_base_url_for_trainee_model \
   TRAINEE_API_KEY=your_api_key_for_trainee_model \
   graphg --output_dir cache
   

Run from Source

1. Install dependencies

   pip install -r requirements.txt
   

2. Configure the environment
   • Create an .env file in the root directory

     cp .env.example .env
     

   • Set the following environment variables:

     # Synthesizer is the model used to construct KG and generate data
     SYNTHESIZER_MODEL=your_synthesizer_model_name
     SYNTHESIZER_BASE_URL=your_base_url_for_synthesizer_model
     SYNTHESIZER_API_KEY=your_api_key_for_synthesizer_model
     # Trainee is the model used to train with the generated data
     TRAINEE_MODEL=your_trainee_model_name
     TRAINEE_BASE_URL=your_base_url_for_trainee_model
     TRAINEE_API_KEY=your_api_key_for_trainee_model
     

3. (Optional) If you want to modify the default generated configuration, you can edit the content of the configs/graphgen_config.yaml file.

   # configs/graphgen_config.yaml
   # Example configuration
   data_type: "raw"
   input_file: "resources/examples/raw_demo.jsonl"
   # more configurations...
   

4. Run the generation script

   bash scripts/generate.sh
   

5. Get the generated data

   ls cache/data/graphgen
   

🏗️ System Architecture

Directory Structure

├── baselines/           # baseline methods
├── cache/               # cache files
│   ├── data/            # generated data
│   ├── logs/            # log files
├── configs/             # configuration files
├── graphgen/            # GraphGen implementation
│   ├── operators/       # operators
│   ├── graphgen.py      # main file
├── models/              # base classes
├── resources/           # static files and examples
├── scripts/             # scripts for running experiments
├── templates/           # prompt templates
├── utils/               # utility functions
├── webui/               # web interface
└── README.md

Workflow

🍀 Acknowledgements

• SiliconCloud Abundant LLM API, some models are free
• LightRAG Simple and efficient graph retrieval solution
• ROGRAG ROGRAG: A Robustly Optimized GraphRAG Framework