torchrun \ --nnodes=1 \ --nproc_per_node=8 \ train.py \ --model SiT-S/2 \ --epochs=800 \ --data-path /home/nvidia/datasets/imagenet-1k/train \ --wandb \ --global-batch-size=1024 \ --run-name 005-SiT-S-2-Linear-velocity-None \ --ckpt /home/nvidia/SiT-Complementary/results/005-SiT-S-2-Linear-velocity-None/checkpoints/0550000.pt # batch_size x 4, lr x 2