model location on windows

[FilipeD]

Hi there!
Am I missing something?

1. May I ask why must the models, most of which weighting over a couple of GBs, be placed under some directory in the venv instead of simply checking what models are already present in hugging face's cache (hub drive:/users/user/.cache/huggingface/hub) folder? I've become sick and tired of junction-linking stuff... I couldn't keep doing it, I've just changed my local venv hub to do just that...
2. Why do transformers keep insisting on going online before checking if the stuff is already available?
3. Why does it abort training when offline? On saving a checkpoint, throws an exception while attempting to get the model's config.json. Guess what happened when leaving it finetuning overnight only to find it had aborted because it found itself offline... :P