Created from a python script I wrote that generates random plots within certain categories, and then creates about 5-15 responses.
Each response length is randomly selected from a small list to keep the responses dynamic. I also make the LLM respond in third person 2/3 times, and in first person 1/3 times (as I have seen this done sometimes as well)
I also have a cleanup step, where I am using another model to clean up the responses (Sometimes sentences are cut off from reaching the maximum… See the full description on the dataset page:
https://huggingface.co/datasets/SuperbEmphasis/Claude-4.0-DeepSeek-R1-RP-SFWish.
