For opensource models like r1 that do thinking in tags, and then respond, its a bother to have to find where the thinking ends, it would be great if we could customize that if x tag appears, make it collapsable so that the output would either only be answer tags or no tags