kabachuha

@patil-suraj Do you have any plans for reverse functionality? I.e. diffusers .bin -> original .ckpt? Your knowledge would help us greatly

Also seems to cause https://github.com/deforum-art/deforum-for-automatic1111-webui/issues/53 in Deforum

I believe the *advanced* users will read the code comment anyway, but it's creating additional barriers for people who will just want to run the testing snippet and be satisfied...

Some users may not be even knowing that their torch version is. To check it they will have to either do some command-line kung-fu or even make a new python...

Hi! Please consider adding it to Latte as well, because there's a plan to make it the main codebase instead of vanilla DiT And both variants may also benefit if...

Another note is that if you're adding Rotary embeddings, you need to remove the sinusoidal ones, to not repeat them

The code is switching away from DiT

Also making a mirror PR to OpenDiT https://github.com/NUS-HPC-AI-Lab/OpenDiT/pull/92

@LinB203 @sennnnn, anyone willing to test?

Well, the entire point of MMDiT is to process *both* text and image embeddings at the same time. The cost may not be as high, if the better structure allows...