docs: add diffusers tutorial draft

[nifleisch]

Description

I added a tutorial that shows the performance of our optimization engine for diffusers pipes. Right now, I only have FLUX as an example, but I already prepared similar plots for SDXL and SD. If you like it, we could use it as the tutorial for the Hugging Face Diffusers documentation.

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[davidberenstein1957]

@nifleisch, looks fantastic and super useful to get a general understanding of optimising diffuser models. Are these intended to be examples or educational tutorials? As discussed last week, I think it would be better to have some dedicated tutorials like #126, so I would propose merging the efforts or creating a section "Advanced Tutorials" but happy to hear your thoughts :)

[nifleisch]

@nifleisch, looks fantastic and super useful to get a general understanding of optimising diffuser models. Are these intended to be examples or educational tutorials? As discussed last week, I think it would be better to have some dedicated tutorials like #126, so I would propose merging the efforts or creating a section "Advanced Tutorials" but happy to hear your thoughts :)

I completely agree—this doesn’t really belong in our educational tutorials. I created this notebook because I believe the algorithms available in Pruna yield really impressive results that we haven’t yet communicated. I made most of the plots in preparation for my talk at our Efficiency Meetup so I thought they would also make a nice tutorial. Perhaps we should publish it as a blog post instead of adding it to our documentation tutorials. Alternatively, we could include it as the Pruna-specific tutorial in the Diffusers optimization docs. Most of their acceleration guides mainly report benchmark results (e.g. https://huggingface.co/docs/diffusers/en/optimization/para_attn), so this tutorial would fit in there nicely.

[sharpenb]

Very nice tutorial! Some comments:

• Add legend for grey lines for compiler and quantizer plots
• Add more methods on diffusers_combination plot
• As part of external docs, the title "Pruna" is good but we should probably cahnge it when putting it in our docs.
• The subtitles "Flux.1 [dev]" are a bit redundant. Let's switch the first subitle for sthg like "Compress Flux.1 [dev]" and remove the others.
• E.g. in "You can choose between 3 cachers", we should make clear that this is for the open-source pruna.
• In each algo family, it would be nice to explicitly show the share structure i.e. "Intuition:", "Code:", "Evaluation:"
• Some closing words with maybe relevant links could be nice at the end of the tutorial.

[davidberenstein1957]

@nifleisch this would also fit nicely here https://github.com/huggingface/cookbook