How to confirm the loss convergence

During reproduction, I noticed that the loss of my dataset converges very slowly. May I ask what the approximate starting and ending points of the loss are under your ViT-G configuration?