Comment by bestcoder69

Then this is SD for Apple silicon users. 13B runs on my m1 air at 200-300ms/token using llama.cpp. Outputs feel like original GPT-3, unlike any of the competitors I’ve tried. Granted- non-scientific first impressions.