I used the Meta-Llama-3-70B-Instruct.Q4_0.llamafile (quantized at Q4) file from here:

→ View original post on X — @simonw