A large model task detection method, device, medium and equipment

Abstract

The embodiment of the specification discloses a large model task detection method, which obtains training samples containing input data of a task and information required for inference of a large model as positive samples, changes the information required for inference in the positive samples, determines negative samples, constructs first samples, obtains second samples containing complete tasks and disassembled sub-tasks, trains a task detection large model through the two kinds of samples, learns the ability of judging whether the disassembled sub-tasks are executable, thereby determining whether there is a situation that the input large model task granularity is too large, and determining the strategy of executing the task. Through the construction of positive and negative samples and task disassembly samples, the model learns the ability of judging the rationality of the task granularity, which reduces the labor cost, and also does not need to set test cases and run the large model, thereby reducing the resource consumption.

Images