Model loading from Huggingface cache

[mpetruc]

Is there a way to load models from the huggingace .cache folder? I'd like to avoid downloading models for both transformers and exllamav2. Also, i'm hoping to speed up model loading (loading Mistral 7b takes ~10 mins with exllamav2+tabbyapi vs ~30 secs with transformers from cache). Thank you.