Comment by teamcubitflow

I'm surprised that kind of captioning came from a 2B model; glad the fine tuning process actually shows a deliberate approach to making qwen 3.5 into essentially a new model of it's kind.