Instructions to use ostris/vae-kl-f8-d16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use ostris/vae-kl-f8-d16 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("ostris/vae-kl-f8-d16", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
why force_upcast=True
#2
by vladmandic - opened
as title says, why force_upcast=True in config? that's only for broken VAEs that overflow in FP16.
I didn’t alter the config. It was just a conversion from LD format. I assume that is the default for diffusers. I will do some testing without upcast at FP16 and change that if it doesn’t overflow.