Welcome to the official repository for the Graph-R1 project. This repository hosts the code, sampled data, and essential resources dedicated to graph-based reasoning in natural language processing (NLP) tasks.
This project explores the use of graph-based structures to boost reasoning capabilities within NLP models. The implemented methodologies and experiments are designed to tackle the challenges of understanding and reasoning over intricate relationships represented as graphs. The trained models can be found in huggingface-hub: https://huggingface.co/finduzzz.
code/: Contains the source code for model SFT training, reinforcement training, and evaluation.data/: Includes sampled datasets and preprocessed graph structures used in the experiments.
For detailed instructions on how to use the code and data, please refer to the README file within the code/ directory.