Paddle Graph Learning (PGL) is a deep learning framework specialized for graph neural networks, built on top of PaddlePaddle. It provides efficient implementations of message passing, sampling, and heterogeneous graph operations. The framework includes domain-specific applications for knowledge graph embedding (Graph4KG), recommender systems (Graph4Rec), and large-scale graph training with GNNAutoScale, achieving top results in OGB-LSC KDD CUP 2021 competitions.