model: mobileface
device: NVIDIA 2080 * 1
dataSet: glint360K
Based on the pre training model, continue training for about 30 days
loss is reduced only in a few iterations after it is reduced, and then it is almost the same