Processing method and processing system for operator output length in AI inference engine

The invention provides a processing method and a processing system for operator output length in an AI inference engine, and relates to the technical field of deep learning models. The method comprises the following steps: firstly, determining a target operator from an initial calculation graph of a deep learning model, then selecting at least two target output lengths from a plurality of output lengths of the target operator, generating a target calculation graph corresponding to each target output length, and then running each target calculation graph by using calculation graph running software. And meanwhile, generating an AI inference engine corresponding to each target calculation graph, finally determining a corresponding target AI inference engine according to each target data in the target data set of the deep learning model, and running the target AI inference engine. According to the technical scheme, the corresponding AI inference engines are generated by selecting the at least two target output le