Comment by ls612

On the smaller end, Quen 3.8, while being extraordinarily capable for a small local model, also suffers from extreme thinking. I wonder if the techniques described here generalize to other models too.