As someone who desperately wants to use local models, I lament there is no way to use them on consumer hardware for serious coding work. I have a rtx 4070 super ti and I cannot run any large model with enough context and tps compared to a remote offering.