This is wild: chatjimmy.ai ~15K tokens/sec 🤯 Custom silicon running Llama 3.1 faster than any GPUs. When this hits newer models, whole new categories of apps unlock.