Inference is much slower with flag --pretrained-model and no internet connection #81
Description
Hello,
I have SF3D running on a VM machine that at inference time won't have access to internet. So i cloned the repo from stabilityai/stable-fast-3d into a local folder and when i run SF3d i set --pretrained-model <local_folder>.
IF I run the model with internet connection and without the flag --pretrained-model <local_folder>, it takes about 20 seconds,
IF I run the model without internet connection and with --pretrained-model <local_folder>, it takes about 2 minutes and 20 seconds.
In addition I also tried to replace pretrained_model_name_or_path: "facebook/dinov2-large" with the path of the local folder where i cloned dinov2-large, but at inference time without internet connection it still takes about 2 minutes.
Am I missing something?
Thankssss