That's completely not true. LLM on device would use MORE electricity. Service providers that do batch>1 inference are a lot more efficient per watt. Local inf

That's completely not true. LLM on device would use MORE electricity. Service providers that do batch>1 inference are a lot more efficient per watt. Local inference can only do batch=1 inference, which is very inefficient.