I downloaded the imageNet dataset in its entirety, trained on DiC-S, and trained 80 epoches. After that, run the sample_ddp.py file and pass in the training results of the 0400000.pt, which is the default value of 1.5 used by cfg-scale. At the same time, I implemented a code to get the .npz file of the imageNet validation set. Running the evaluator.py file, we get an IS value of 50.63 and a FID value of 27.66. But the results of your paper (Part 4.2) on DiC-S are: IS value of 25.82 and FID value of 58.68. Why is this happening? Am I doing something wrong? Looking forward to hearing from you.