Viper: A High-Performance I/O Framework for Transparently Updating, Storing, and Transferring Deep Neural Network Models
An I/O framework that determines near-optimal checkpoint schedules and accelerates delivery of the latest model updates for DNN training and inference.
read more →