[docs] Performance tutorial #2773
Conversation
stevhliu
requested review from
patrickvonplaten,
pcuenca,
sayakpaul and
yiyixuxu
|
The documentation is not available anymore as the PR was closed or merged. |
docs/source/en/_toctree.yml
Outdated
| title: Overview | ||
| - local: using-diffusers/write_own_pipeline | ||
| title: Understanding models and schedulers | ||
| - local: stable_diffusion |
There was a problem hiding this comment.
Shouldn't we give a bit less generic name to this doc since we're discussing optimization specifically?
There was a problem hiding this comment.
@sayakpaul I think that's just the file name, the name of the doc is "Improve DiffusionPipeline performance" -
There was a problem hiding this comment.
Yeah and this way we don't break any links to this doc!
There was a problem hiding this comment.
I'd actually like to keep it a main doc page for now
There was a problem hiding this comment.
Thanks for the updates! I really like the isolation of the information.
@patrickvonplaten should we consider updating this doc with the latest goodies we support from diffusers:
- VAE tiling (for a higher-resolution generation without blowing up memory)
- CPU offloading
- UniPCMultistepScheduler (fastest scheduler)
docs/source/en/_toctree.yml
Outdated
| title: Overview | ||
| - local: using-diffusers/write_own_pipeline | ||
| title: Understanding models and schedulers | ||
| - local: stable_diffusion |
There was a problem hiding this comment.
@sayakpaul I think that's just the file name, the name of the doc is "Improve DiffusionPipeline performance" -
| ``` | ||
| ```python | ||
| pipeline = pipeline.to("cuda") | ||
| ``` |
There was a problem hiding this comment.
should we update this guide with enable_model_cpu_offload()? cc @patrickvonplaten
* update performance tutorial * fix divs * oops forgot to close tag * apply feedback * apply feedback * apply feedback * align doc title
* update performance tutorial * fix divs * oops forgot to close tag * apply feedback * apply feedback * apply feedback * align doc title
5 participants
This PR updates the Stable Diffusion guide to Improve DiffusionPipeline performance tutorial. Some of the updates include:
StableDiffusionPipelinefor the more genericDiffusionPipelineDiffusionPipelinefirst before trying to optimize it[[open-in-colab]]tag and removes the link to the Colab notebook