2026-04-17T02:49:53Zhttps://recercat.cat/oai/request

oai:recercat.cat:2117/3566352025-07-17T14:07:57Zcom_2072_1033col_2072_452950

urn:hdl:2117/356635 An oracle for guiding large-scale model/hybrid parallel training of convolutional neural networks Njoroge Kahira, Albert Nguyen, Truong Thao Bautista Gomez, Leonardo Takano, Ryousei Badia Sala, Rosa Maria Wahib, Mohamed Àrees temàtiques de la UPC::Informàtica::Arquitectura de computadors::Arquitectures paral·leles Deep learning Neural networks (Computer science) Parallel processing (Electronic computers) Model parallelism Performance modeling Aprenentatge profund Xarxes neuronals (Informàtica) Processament en paral·lel (Ordinadors) Deep Neural Network (DNN) frameworks use distributed training to enable faster time to convergence and alleviate memory capacity limitations when training large models and/or using high dimension inputs. With the steady increase in datasets and model sizes, model/hybrid parallelism is deemed to have an important role in the future of distributed training of DNNs. We analyze the compute, communication, and memory requirements of Convolutional Neural Networks (CNNs) to understand the trade-offs between different parallelism approaches on performance and scalability. We leverage our model-driven analysis to be the basis for an oracle utility which can help in detecting the limitations and bottlenecks of different parallelism approaches at scale. We evaluate the oracle on six parallelization strategies, with four CNN models and multiple datasets (2D and 3D), on up to 1024 GPUs. The results demonstrate that the oracle has an average accuracy of about 86.74% when compared to empirical results, and as high as 97.57% for. Peer Reviewed Postprint (published version) 2021 Conference lecture Open Access European Network of Excellence on High Performance and Embedded Architecture and Compilation (HiPEAC)