The invention discloses a compiling optimization method and device for a machine learning model. The method comprises the steps of operating a first machine learning model multiple times to process aset of preheating data, and model operation information of the processing process is collected; according to the model operation information, determining the operation with the tensor shape unchangedin the first group of operations as a target operation; and under the condition that the target operation comprises a plurality of operations in the first group of operations, combining the target operations in the first machine learning model through the target compiler to obtain a second machine learning model. According to the method and the device, the technical problems of long compiling period and resource waste caused by the fact that sampling information during operation is not introduced during compiling of machine learning are solved.