The invention provides a cross-machine learning platform model definition protocol and configuration system, and relates to the field of artificial intelligence automatic machine learning. The systemcomprises a computational graph construction engine; a computational graph optimization engine; a cross-machine learning platform model definition protocol (Cross-Platforms Model definition Protocol);a computational graph encoding engine; a hyper-parameter recommendation engine; a protocol parsing engine; a model training process control assembly; a hardware information detection and configuration system. The system has the advantages that cross-platform machine learning network construction is achieved, the system is easy to use, and the machine learning maintenance threshold is lowered; machine learning network construction in a dragging mode is achieved, and the latest algorithm network can be iterated rapidly; a large number of algorithm models are provided, and machine learning and deep learning network construction under different scenes are supported; a parameter recommendation function is provided, and a lot of parameter adjustment time is saved; algorithm network constructionfrom data cleaning to model release full life cycle is provided, and the system can be directly put into production and use.