METHOD FOR TRAINING NEURAL NETWORK AND RELATED DEVICE

A method for training a neural network and a related device. The method can be used in a scenario of training a neural network in the field of artificial intelligence. The method comprises: when executing an Nth round of training of a neural network, acquiring a first computational graph, wherein the first computational graph is one of one or more computational graphs corresponding to the Nth round of training of the neural network; and after determining that a first compiled code corresponding to the first computational graph is stored in a system, directly executing the first compiled code, wherein the first compiled code is generated during execution of an Mth round of training of the neural network, and M is smaller than N. Since there is no need to execute the operations of converting the first computational graph into intermediate computation representation, and obtaining a compiled code on the basis of the intermediate computation representation, the overhead of computer resources is reduced. La présen