gradio demo bug

[MontaEllis]

I modify the local model fold

but got

[zhengzangw]

I add a force_huggingface in the model config and thus you can use locally download hugging face checkpoint.

[zhengzangw]

An update: now we support loading both the diffusion model and vae from your local path. Otherwise, if you want to load saved checkpoints from training (the previously default behaviour), set

model = dict(
  ...
  force_huggingface = False,
)