Metainformation calculation graph generation system

The invention discloses a meta-information computational graph generation system, and relates to the technical field of deep learning, in particular to a meta-information computational graph generation system which is characterized by comprising a meta-information agent, a tracker with meta-information computing capacity, patch operation and a meta-information computational graph. According to the meta-information computational graph generation system, through combined use of the meta-information agent, the tracker with the meta-information computing capacity, the patch operation and the computational graph of the meta-information, the agent has more pieces of information such as shapes, types and dimensions. Meanwhile, by hijacking an operator calculation function in PyTorch, the method has the capability of flowing in forward propagation of the model like a normal tensor, has high universality, supports an immediate mode frame and control flow tracking, is high in calculation graph generation speed and low