inference script just reconstructs training images

[jhurlbut]

I was able to successfully train a model after chaning the diffusers version and changing batch size to 2 but when running inference on the output I only get reconstructions of the training images

[Victarry]

The reason may be the training iterations are too much,
you can decrease the iterations by reducing the number of sampled images or early stopping.
Generally about ~200 iterations are ok.