I see two big constraints to use it right now: - This runs on the CPU only and does not require GPU. It requires around 60GB of CPU memory for Vicuna-13B. als

I see two big constraints to use it right now: - This runs on the CPU only and does not require GPU. It requires around 60GB of CPU memory for Vicuna-13B. also - We release Vicuna weights as delta weights to comply with the LLaMA model license. You can add our delta to the original LLaMA weights to obtain the Vicuna weights. Instructions: ... This conversion command needs around 60 GB of CPU RAM. Hope they can optimize it further, to make it work with lesser hardware.