[cyanydeez]

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

[wmf]

OK, B70.