PaddleBox: Communication-Efficient TeraByte-Scale Model Training Framework for Online Advertising

Click-through rate (CTR) prediction is one of the most crucial components in the online advertising industry. In order to produce a personalized CTR prediction, an industry-level CTR prediction model commonly takes a high-dimensional (∼ 10
¹²
) sparse vector (that is encoded from query keywords, user portraits, etc.) as input. As a result, the model requires Terabyte scale parameters to embed the high-dimensional input. Hierarchical distributed GPU parameter server has been developed at Baidu to enable GPU with limited memory to train the massive network by leveraging CPU main memory and SSDs as secondary storage. In this work, we identify two major challenges in the existing GPU training framework for massive-scale ad models and propose a collection of optimizations to tackle these challenges: (a) the GPU, CPU, SSD rapidly communicate with each other during the training. The connections between GPUs and CPUs are non-uniform due to the hardware topology. The data communication route should be optimized according to the hardware topology; (b) GPUs in different computing nodes frequently communicates to synchronize parameters. It is thus required to optimize the communications so that the distributed system can become scalable. In this paper, we propose a hardware-aware training workflow that couples the hardware topology into the algorithm design. To reduce the extensive communication between computing nodes, we introduce a k-step model merging algorithm for Adam and provide its convergence rate in non-convex optimization. To the best of our knowledge, this is the first application of k-step adaptive optimization method in industrial CTR model training. Experiments on commercial search ads data confirm the effectiveness of our proposed training framework.