Fix configuring VAE from single file mixin

[DN6]

What does this PR do?

The scaling factor isn't correctly set when calling AutoencoderKL.from_single_file for SDXL models. It defaults to the SD scaling factor, leading to a washed out effect on the final image.

Additionally SD.Next uses the config_file argument to pass in the path to the original config file, rather than the original_config_file argument. This is missing in the current Mixin.

This PR:

1. Allows setting VAE scaling_factor from the VAESingleFileMixin
2. Update so that AutoencoderKL.from_single_file can accept a config_file argument as well.

Fixes # (issue)
#6923

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.

[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.

[sayakpaul]

Agree with @patrickvonplaten's comment.

[yiyixuxu]

hey thanks!
feel free to merge if @sayakpaul and @patrickvonplaten gives ok
I have a few questions (they might be very obvious)

1. is it that the only way to load it correctly we would need to pass the scale_factor manually?

vae = AutoencoderKL.from_single_file("https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/blob/main/sdxl.vae.safetensors", scaling_factor=0.13025)

2. is it not possible to infer the correct config based on the checkpoint in infer_original_config_file?

[sayakpaul]

I think it's because we cannot assume that the scaling factor will always be present in the single-file checkpoint format.

[yiyixuxu]

@sayakpaul
ohh infer_original_config_file returns a URL to the config file, currently it returns the v1 https://raw.githubusercontent.com/CompVis/stable-diffusion/main/configs/stable-diffusion/v1-inference.yaml

I'm just wondering if it's able to know it's sdxl and return correct config URL instead

[yiyixuxu]

also which one is correct config file for this vae, is it this? https://github.com/Stability-AI/generative-models/blob/main/configs/inference/sd_xl_base.yaml

[DN6]

ohh infer_original_config_file returns a URL to the config file, currently it returns the v1 https://raw.githubusercontent.com/CompVis/stable-diffusion/main/configs/stable-diffusion/v1-inference.yaml

I'm just wondering if it's able to know it's sdxl and return correct config URL instead

I think we can diff the state_dicts between SD 1.5, SD 2 and SDXL to see if there are unique keys for the model type that we could use to infer the config. Essentially in this case, we only have the VAE checkpoint state dict to make our guess. For the pipelines calling from_single_file we rely on using the TextEncoder and UNet to infer the model type.

I can open a follow up PR to look into it.