3 ms·
You can run the full Deepseek 671b q4 model at 40 tokens/s. 37B active params at a time because R1 is MoE.
by aurareturn 2y ago
You can run the full Deepseek 671b q4 model at 40 tokens/s. 37B active params at a time because R1 is MoE.
- KingOfCoders 2y agoIn another of your comments it was "by my calculation". Now it's just fact?
- aurareturn 2y agoEasiest calculation.
- KingOfCoders 2y agoMe pulling numbers out of my hat, easiest calculation too.
- aurareturn 2y agoTroll.
- KingOfCoders 2y agoExactly. Sub-species Apple fanboy.
- aurareturn 2y agoWannabe CTO. Engineering manager at some random Wordpress real estate website and an ex-CTO at an "eBay subsidiary". Looks to be an misinformed AMD fanboy as well. Trolls HN trying to advertise a book that was probably written by ChatGPT.
- KingOfCoders 2y ago[flagged]
- aurareturn 2y agoYou ain't owning anything. I've seen countless "coaches" like yourself with questionable experience. Looks more like a BS artist honestly.