Task-oriented training method and use method of generative large language model

Abstract

This disclosure provides a method for training and using a generative large language model oriented towards task solving, relating to artificial intelligence technologies such as generative large language models and unsupervised training. The method includes: training a general generative large language model based on a first training sample constructed from different tool applications and corresponding functional description documents, obtaining a first model; in a controllable operating environment integrating various tool applications, controlling the first model to automatically explore and invoke multiple tool applications to complete the user needs represented by the preset user input, obtaining the actual execution results output by the first model; updating the first model using reinforcement learning based on the similarity between the standard execution results and the actual execution results, obtaining a second model; training the second model using reinforcement learning based on human feedback, thereby obtaining a target generative large language model that can accurately invoke tool applications to solve user tasks.

Images