Impact of batchsize on performance during sampling

[return-sleep]

Thanks for your nice work. I tried this method on my own dataset and observed a phenomenon. During the testing phase, the performance of the model is affected by the batch size. It seems that a small batch size will give better results, but this will increase the time spent evaluating on the test set. Is this reasonable, or is there any way to fix it?

[return-sleep]

I guess it may be caused by ConditioningMethod.grad_and_value (here), where batch size influences the gradient norm and then influences the guidance step. Another question is how can we replace the DDPM sampling method with DDIM to speed up the inference with the same guidance? As the number of steps sampled decreases, the step size looks like it needs to be increased accordingly to achieve a similar result.

Hi, how do you set the batchsize in the code?