What local LLM-s you use?

[Rover4444]

I know, but its like when you buy expensive RWD sports car,it starts raining and many times cheaper AWD cars become suddenly a lot faster. You know your fancy sports car is not meant to be fast in rain, but for the money* you paid for it, you wan't it to be.

It's more like bringing a Jeep to a racetrack and expecting it to beat a supercar. The Jeep is built for off-roading, after all - something a supercar can't do at all.

Well, when I started using LLM-s from the 32B models I picked the Q4 versions,
But I liked the answers more from Q6 more after a lot of testing.
Also Bartowski recommending a lot of quants, including Q6 with the note very high quality
View attachment 388121
Would be nice to have an 5090 to run this from VRAM ofc.
Some day we will!

If I need a quick answer, I have smaller models ready to load!

I'll take your word for it. I guess we just use our systems differently.

 

[csendesmark]

It's more like bringing a Jeep to a racetrack and expecting it to beat a supercar. The Jeep is built for off-roading, after all - something a supercar can't do at all.

I'll take your word for it. I guess we just use our systems differently.

Just completed testing with the new driver, also that previous number came up with doing other browser stuff and watching YouTube
LM Studion only gave this:

6.29 tks is more than double