[Feat]: Enable xformers efficient attention for performing video super resolution in a limited VRAM environment

Vchitect / LaVie

LaVie: High-Quality Video Generation with Cascaded Latent Diffusion Models

Apache License 2.0

852 stars 60 forks source link

[Feat]: Enable xformers efficient attention for performing video super resolution in a limited VRAM environment #67

Open kimx3966 opened 1 month ago

kimx3966 commented 1 month ago

What does this PR do?

This PR enables xformers efficient attention for performing video super resolution in a limited VRAM environment. Additionally, it enables half precision inference for the VAE model.