sidney1505

I'm also very interested in the training code!

Hi, the training process appears to be a bit numerically unstable and nan numbers can happen - for me it helped to use smaller learning rates. However, I also did...

I tried scaling it up and it kind of works, but needs some tuning depending on the dataset. However, I encounter some problems, e.g. that the flow is not perfectly...

I still experience the same problem

does the code base work for you for unconditional models and can you reproduce the results?

@ozgurkara99 did you find a solution for the problem? I tried it the other way around doing inference with the pretrained models and that also did not work...

does the code base work for you beside this and you can reproduce the results?

@hycsy2019 did you find a reason for this? I have very similar issues to you...