From #4428 (comment), the Optimization/Special Hardware section has grown and it may be time to reorganize it. I want to explore how we can do this, and my initial thoughts are:
To me, there are three subtopics in this section that we can reorganize on:
- General optimizations that you can apply to
DiffusionPipeline (memory/speed, torch 2.0 support, xFormers, token merging)
- Optimized or specialized model types (
OVStableDiffusionPipeline, ORTStableDiffusionPipeline, CoreML models, Stable Diffusion in JAX/Flax)
- Hardware (MPS, Habana)
For the "Memory/speed" page, I think we can even split this into two docs; one focused on saving memory and the other on speeding up inference.
Happy to hear your thoughts! cc @pcuenca @patrickvonplaten @sayakpaul @yiyixuxu
From #4428 (comment), the
Optimization/Special Hardwaresection has grown and it may be time to reorganize it. I want to explore how we can do this, and my initial thoughts are:To me, there are three subtopics in this section that we can reorganize on:
DiffusionPipeline(memory/speed, torch 2.0 support, xFormers, token merging)OVStableDiffusionPipeline,ORTStableDiffusionPipeline, CoreML models, Stable Diffusion in JAX/Flax)For the "Memory/speed" page, I think we can even split this into two docs; one focused on saving memory and the other on speeding up inference.
Happy to hear your thoughts! cc @pcuenca @patrickvonplaten @sayakpaul @yiyixuxu