According to Tianyancha App, on February 7th, a patent for "training methods, devices, Computer Equipment, and storage media for large language models" applied by Tencent Technology (Shenzhen) Co., Ltd. was published. The abstract shows that in this method, by introducing a first summary text and a second summary text during the training process of the large language model, more learnable information is provided for model training. At the same time, since the amount of information in the first and second summary texts is different, and the first summary text contains correct and incorrect statements, comparing the two different summary texts of the same base text and distinguishing the correct and incorrect statements in the first summary text avoids problems such as model overfitting and inaccurate generation caused by the singular nature of summary texts. This not only improves the generalization performance of the model but also enhances the accuracy of the model.