The performance depends on the hardware you use. If you have access to more powerful hardware, you can run 13B and 70B models by following the instructions here: https://github.com/getumbrel/llama-gpt