i saw the time stamps. 1 day ago like 12th? then unsloth release the quants about half a day later.
so i need to link like 25 strix halos? 
Isn’t it active-95b-tokens so you might survive with 128GB RAM and all that swapping-like access from a 4TB NVMe?
But yeesh, that’s a big lad.
i only have a lousy 10Mbps connection. just downloading will take a loooooonng time . but having already tried qwen 122B and oss 120B, i must say being able to get 2.4T going would persuade many to ditch paid AI services. 
Maybe you can get some straight from ssd action going and maybe get a few tokens/minute XD
1 Like
toks/minute ! i will try it on my pentium iii 500Mhz !
Tokens/century then I guess.
Gonna be interesting how the 27B version that should be released today performs, weirdly this week seems to be the “hey everyone release new 30ishB open weight models”-week.
1 Like