SuperbEmphasis's picture
Update README.md
9b8b9e8 verified
metadata
size_categories:
  - 1K<n<10K
extra_gated_prompt: >-
  This Dataset contains sexual explicit material.  Please confirm your age by
  checking the box below.
extra_gated_button_content: Age Restriction
extra_gated_fields:
  I confirm that I am over the age of 18: checkbox

Dataset ERP dataset generated from deepseek R1. This was a bit intensive. I write a python scrip that would generate a random plot, the ERP theme of which was chosing at random from about 20 different types of themes (One of the more tame ones, for example, "Romance Erotica"). I would then guide the inference by using the "user" role to switch perspectives between the two characters, and continue the story.

I then wrote another script that used deepseek-cat to fix the reasoning portions. Since I my instructions were automated, but I would frequently get a "Now the user wants me to switch roles..." Types of thinking. So I used deepseek-chat to clean up the reasoning.

Lastly, I reprocess the conversation, removing my "user" instruction. This left ALL of the conversation as the "assistant" role. So I iterated through this and alternated between user and assistant.

All of these conversations were generated using thinking.

Ultimately this has about 1300 conversations, all of which are 5-15 turns! I might try to add more later that are longer (I.E 15-30+ turns).

I did have a rule I added to all of the system prompts though. I mandated that all characters are 18 years old, or older.