puwaer/Doujinshi-14b-roleplay
puwaer/Doujinshi-14b-roleplay is a 14 billion parameter large language model, based on Qwen/Qwen3-14B, specifically fine-tuned for R18-themed roleplay. It was trained using 4 billion tokens of R18-specific Japanese datasets scraped from dmm.co.jp and dlsite.com, and supports a context length of 32768 tokens. This model excels at maintaining character persona, tone, and first-person perspective in interactive roleplay scenarios, making it suitable for immersive conversational experiences.
Loading preview...
Model Overview
puwaer/Doujinshi-14b-roleplay is a 14 billion parameter large language model (LLM) built upon the Qwen/Qwen3-14B architecture. It has undergone continuous pre-training, DPO (Direct Preference Optimization), and SFT (Supervised Fine-Tuning) specifically for R18 content.
Key Capabilities
- Specialized Roleplay: This version is uniquely fine-tuned for character-driven roleplay, excelling at maintaining a consistent persona, tone, and first-person perspective throughout conversations.
- R18 Content Generation: Trained on 4 billion tokens of R18-specific Japanese datasets from dmm.co.jp and dlsite.com, it is designed to generate content suitable for adult-themed scenarios.
- Immersive Interactions: Facilitates engaging and immersive conversational experiences by adhering to character settings and dialogue styles.
Training Data
The model was trained using a comprehensive set of R18-focused Japanese datasets, including:
- puwaer/dlsite-jp-v1
- puwaer/dlsite-jp-v2
- puwaer/dlsite-jp-v3
- puwaer/dmm-fanza-jp-v1
- puwaer/dmm-fanza-jp-v2
- puwaer/dmm-fanza-jp-v3
- puwaer/Doujinshi-sft-dataset-v1
- puwaer/Doujinshi-dpo-dataset-v1
Good For
- Creating interactive R18 roleplay scenarios.
- Applications requiring an LLM to consistently embody a specific character's personality and speaking style.
- Generating immersive and detailed adult-themed narratives and dialogues.