not at the vram sizes that control how much context to load; also, GPUs arn't as effiecient as direct inference.