Jonathan Whitaker

> I'll be working on this soon. I've been chatting with the Quarto folks about this, and they have created a paved path that will make this possible! More to...

This is especially useful for things like progressive distillation (https://arxiv.org/abs/2202.00512) where they found predicting the noise (epsilon) had some issues, and instead recommend either: - Predicting x directly - Predicting...

There will have to be some additional logic in all of the samplers to handle models with different objectives, but given any one of the three listed you can derive...

I think I'm on team 'three classes'. If you want to do a quick demo of one application, easy to just load, say, text2img. If you want multiple your example...

Thanks for sharing a solution. Removing '-arch compute_30' let me install on Google Colab.

The model in diffusers expects a timestep as the second argument but since we're training from scratch we can choose to ignore it by always passing 0 as the timestep....

I edited some schedulers to expose that a while back, I'll see how easy it is to do the same for PNDMScheduler, but that could take a while :)

If you try to access a model on the hub that requires you to be logged in, you need to log in. That first error is saying that it can't...

I think on some shared machines `export CUDA_VISIBLE_DEVICES=1,0` might reference cards other than the ones you're assigned. (Don't quote me on this but I think I just hit a similar...

@lewtun what do you think? I hadn't merged because 1) not sure we want all the badges there (maybe at the end instead where they'll break the flow less) and...