5,000 synthetic Japanese roleplay dialogues generated via OpenRouter's gpt-5-chat, featuring approximately 20 turns per interaction. The data includes structured metadata for world-building and character personas, specifically targeting NSFW roleplay scenarios.
Use Cases
- Fine-tune LLMs for Japanese roleplay capabilities using the system_message and conversations fields
- Develop content moderation tools for NSFW Japanese text using the tag and minor_genre labels
- Study stylistic variations in Japanese dialogue by analyzing the dialogue_tone column relative to the conversation text
- Generate character-driven responses by conditioning models on the assistant_setting and world_setting metadata
Strengths
- 5,000 dialogue samples with an average length of 20 turns per conversation
- Categorized by major_genre and minor_genre for specific roleplay themes
- Includes explicit character and scene definitions via user_setting, assistant_setting, and scene_setting columns
- Contains a tag field specifically for R-18 age-restricted content identification