This is an experiment. For RP, sometimes the thinking that Deepseek and others do are a bit much. Since I Now have a rather large (ish) deepseek R1 dataset. I made a python script to count the number of words from the reasoning, round up to the nearest 50 (So if the number was 416, this would be set to 450), and then add some text in the system prompt and the reasoning.
My goal is to easily, via a system prompt, be able to control the size/amount of tokens from thinking.
