Comment by nicce

You can run it with 2x r9700 with 150-200 tokens per second. It is intelligent enough if you just point the docs / whatever for it.