https://mega.nz/folder/GZ8DnBjZ#oc70hM9RGslozRhZDQJkwA

According to the number of training times, it is divided into five versions: 2, 4, 6, 8, and 10. The overfitting of 8 and 10 is strong but the composition is relatively stable. However, it needs to be matched with i2i high-resolution correction to produce pictures. For those who don’t care about the beheading effect, using versions 2, 4, and 6 should be enough.

The main difference between the open version of LORAv2c in this training is the classification of four concepts.