Generative Flow Networks (GFlowNets; Bengio et al., 2021) is a novel learning framework that has shown promising results in matching unnormalized reward distributions. Unlike traditional reinforcement learning methods, GFlowNets can achieve reward matching and discover high-quality and diverse solutions, making it useful for various applications such as drug discovery, recommender systems, and dialog systems. However, GFlowNets suffers from low learning efficiency, which limits its practicality and scalability. This project aims to address the low learning efficiency of GFlowNets and improve its training performance.