Fix image upcasting

[tolgacangoz]

Thanks for the opportunity to fix #7854

What does this PR do?

This PR proposes to fix image upcasting before vae.encode() when using fp16 and vae.config.force_upcast==True with xformers or torch>=2.0 installed. Casting with .to(next(iter(self.vae.post_quant_conv.parameters())).dtype) is supposed to be preferred before vae.decode() not vae.encode().

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you read our philosophy doc (important for complex PRs)?
• Was this discussed/approved via a GitHub issue or the forum? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag members/contributors who may be interested in your PR.
@yiyixuxu @kadirnar

[bghira]

the same issue exists in the StableDiffusionPipeline :) would you like to tackle that one too? latents must be cast to vae dtype during decode

[tolgacangoz]

I need to understand. Wouldn't this throw an error:

import torch
from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5",
                           torch_dtype=torch.float16, variant='fp16').to("cuda")
image = pipe("a photo of an astronaut riding a horse on mars").images[0]

Or, is it a case in MPS?

[bghira]

it's actually something that occurs on ROCm which masquerades as CUDA

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[yiyixuxu]

thanks!

[tolgacangoz]

Thanks for merging!

[kadirnar]

Thanks @StandardAI ❤️