Code defect detection model training method, device and system

The invention provides a code defect detection model training method, device and system. The method comprises the steps that a first code graph is generated based on a first source code, and the first code graph is used for reflecting the relation between objects in the first source code; a first variable quantum graph is generated according to a first code graph, the first variable quantum graph is a relationship among partial objects extracted from the first code graph based on a first variable, and the first variable is a partial object in the first source code; detecting the first variable quantum diagram to determine the label of the first variable quantum diagram; and training a code defect detection model by using the first variable quantum graph and the label of the first variable quantum graph. According to the method, the memory code defects can be quickly detected. 本申请提供了一种代码缺陷检测模型训练方法、设备以及系统。所述方法包括：基于第一源代码生成第一代码图谱，其中，第一代码图谱用于反映所述第一源代码中的对象之间的关系；根据第一代码图谱生成第一变量子图，其中，所述第一变量子图是基于第一变量从所述第一代码图谱中提取出的部分对象之