LARGE DEEP LEARNING MODEL TRAINING METHOD AND SYSTEM, DEVICE, AND MEDIUM

摘要

Disclosed in the present invention are a large deep learning model training method and system, a device, and a storage medium. The method comprises performing the following steps on each topological layer: arranging tensors in an ascending order according to series numbers of required topological layer levels of the tensors; sequentially carrying the tensors to a GPU according to the arrangement, and determining whether the sum of the tensors already carried to the GPU exceeds a threshold; in response to the fact that the sum of the tensors already carried to the GPU exceeds the threshold, carrying the excess part to a CPU, and determining whether the current topological layer is the last topological layer; and in response to the fact that the current topological layer is the last topological layer, correcting the tensor having an abnormal position. According to the large deep learning model training method and system, the device, and the medium provided in the present invention, a more precise and accurate carrying strategy is formulated depending on a precedence relationship of using the tensors, thereby ensuring maximization of performance.