Currently, there is a hack in place to force the model to download weights upon startup - this involves running the model on a dummy input. Need to figure out if theres a way we can cache those weights and bake it into the image so we don't have to install it every time on the first run.
Currently, there is a hack in place to force the model to download weights upon startup - this involves running the model on a dummy input. Need to figure out if theres a way we can cache those weights and bake it into the image so we don't have to install it every time on the first run.