Comment by zozbot234

Cloud hardware is not inherently more "proper" than what's being proposed here, there's nothing wrong per se about targeting slower inference speeds in an on prem single-user context.