SDXL Turbo support and example launch

[bram-w]

What does this PR do?

I added an --is_turbo argument to the new Diffusion-DPO scripts to support DPO tuning of the 4-step SDXL Turbo model. I also added an example call in the corresponding README

Before submitting

• [ x] Did you read the contributor guideline?
• [x ] Did you read our philosophy doc (important for complex PRs)?
• [ x] Was this discussed/approved via a GitHub issue or the forum? Please add a link to it if that's the case.
  No, but direct follow-up to [Experimental] Diffusion LoRA DPO training #6422
• [x ] Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
  Added readme launch and in-line comments
• [ x] Did you write any new necessary tests?
  N/A

Who can review?

@sayakpaul

[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]

Possible to maybe add a note that it enables 4-steps sampling?

[sayakpaul]

nit: if we remove the help statement out of the brackets and remove the comma it probably will go to the line above.

[sayakpaul]

Can this be explained in the comments a bit? I know about the 4-step part that was discussed in the ADD paper for training of the student.

[sayakpaul]

Awesome! Thanks much.

Just some nits.

[radames]

hi @bram-w could we also have this setup for SD-turbo version? which is based on SD2.1, thanks

[sayakpaul]

@bram-w let me know if you need any help :)

[radames]

@bram-w testing your script you might also need to change the guidance scale and num steps on the validation step, you could add these lines

   guidance_scale = 5.0
    num_inference_steps = 25
    if args.is_turbo:
        guidance_scale = 0.0
        num_inference_steps = 4

and add the params here

diffusers/examples/research_projects/diffusion_dpo/train_diffusion_dpo_sdxl.py

Lines 120 to 124 in dc7f024

	for prompt in VALIDATION_PROMPTS:
	with context:
	image = pipeline(prompt, num_inference_steps=25, generator=generator).images[0]
	images.append(image)

and here

diffusers/examples/research_projects/diffusion_dpo/train_diffusion_dpo_sdxl.py

Line 144 in dc7f024

pipeline(prompt, num_inference_steps=25, generator=generator).images[0] for prompt in VALIDATION_PROMPTS

wdyt @sayakpaul

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[yiyixuxu]

any update on this PR?

[sayakpaul]

@bram-w this PR is already in pretty good state. Let us know if and how we can help to get this merged.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[sayakpaul]

@bram-w a gentle ping :)

[sayakpaul]

@radames do you want to push the changes you suggested to the script and we can then merge?

[radames]

@radames do you want to push the changes you suggested to the script and we can then merge?

yes happy to do that, do I have permission to send commits here?

[sayakpaul]

I think you should be able to since you are a member of the HF org.

[radames]

done @sayakpaul

[sayakpaul]

Thanks a bunch!