What parameters are associated with the training time？

[pokameng]

I have set the diffusion_step=1000,but the progress has run 10000 steps not stopping. My data_dir has 30000 rgb images. There are my settings：

MODEL_FLAGS="--image_size 256 --num_channels 256 --num_res_blocks 2 --learn_sigma True --class_cond False" DIFFUSION_FLAGS="--diffusion_steps 1000 --noise_schedule linear --use_kl False" TRAIN_FLAGS="--lr 1e-4 --batch_size 1 " @longouyang @welinder @JoshuaGross @esigler @jietang

[PatrickSVM]

I would be interested in the same. I guess 30000 rgb images is a lot so of course it will take a lot of time to train. But is there a way to get an estimate for the training progress? The output (among others) prints the amount of steps. I am trying to replicate the unconditional CIFAR10 example with the same hyperparameters you suggested. How many steps will the model take in total? Right now I'm at step 45.000 after a few hours.

[bala1144]

@PatrickSVM What is the config of the machine, you were training it on ?