WOW! This is FAST!
#1
by itsmebcc - opened
I use Qwen/Qwen3.6-27B as a primary coding model, and this is by far the fastest option I have found. Running on an RTX Pro 6000 using standard MTP, I cna get about 80 tok/s. I trained a DSpark model on my coding projects and that did about 140 tok/s. With this setup depending on what I am doing I get 170 tok/s on average.
This does take up some additional space so I cannot run as many concurrent, but it is very fast!
Good job!
Thanks! Great to hear that