AYU_Coserbeta

2034 datasets were used to train the XLlora model for 130,000 steps. Welcome to test the results, currently optimizing the iterations;