Performance at large bottleneck

[MoayedHajiAli]

Thank you for your work.

When training with a DiT that has larger bottleneck and hidden dim that is larger than the patch dim (e.g DiT-XL on ImageNet 256px), did you observe if x0-pred also work better than v-pred?

Thank you!