IQ2_XXS is excellent on Strix Halo
#4
by Cortex0833 - opened
I'm using it to help write prompts with very difficult, dense logic. It's much better than any other model I can run. This includes Qwen3.5 27B on another machine, higher quants of Qwen3.5 122B, various versions of Qwen3 235B.
I should probably note that I always disable thinking because it takes too far too long and the results are only marginally better. My experiences are limited to instruct usage.