What's the point of running it locally though? Inference for open models is quite cheap already. They could just selfhost, anyway. The experience of running LLMs locally will be excruciatingly bad in comparison at least for the near future.