The Local Search GFlowNets is a new training algorithm designed to enhance the sampling quality of GFlowNets by utilizing local search methods. GFlowNets, a training approach for generating structured sequences in a constructive policy, are applicable to various discovery tasks, including drug discovery. They aim to learn a compositional policy proportional to the rewards of the structure. However, the vast combinatorial space reqften leading to an oversampling of low-reward areas. The local search method, which integuired for GFlowNets makes it challenging to learn effectively in highly rewarded regions, orates local backtracking and trajectory reconstruction through the backward and forward policies of GFlowNets, helps target high-reward regions more effectively. By incorporating local search into the exploration phase, improved sampling quality can be achieved across multiple discovery tasks.