Ilya626/gemma-4-31B-it-SDFT-Heretic-RP
Ilya626/gemma-4-31B-it-SDFT-Heretic-RP is a 31 billion parameter Gemma 4 instruction-tuned model, based on the uncensored Heretic SDFT line, and further fine-tuned for enhanced roleplay (RP) behavior. It features stronger scene and character continuity, active narrative pressure, and reduced generic LLM phrasing, making it suitable for dark fantasy, cyberpunk, and grounded RPG narration with a 32768 token context length.
Loading preview...
Model Overview
Ilya626/gemma-4-31B-it-SDFT-Heretic-RP is a 31 billion parameter model built upon google/gemma-4-31B-it. This version incorporates merged Hugging Face weights and is specifically fine-tuned from the uncensored Heretic SDFT base, with a strong emphasis on improving roleplay (RP) capabilities. It is intended for local inference, RP evaluation, narrative testing, and research, offering a 32768 token context length.
Key RP Improvements
This model demonstrates statistically significant enhancements in roleplay-oriented evaluations, including:
- Better scene continuation and handling of multi-turn RP context.
- Stronger character and world continuity, maintaining established traits and active NPC momentum.
- Reduced generic phrasing and less mechanical repetition, leading to more consistent atmospheric detail.
Recommended Use Cases
This model is particularly well-suited for:
- Dark fantasy, cyberpunk, and grounded RPG narration.
- Game Master (GM) style scene continuation.
- Character dialogue and NPC interaction in narrative settings.
- Multi-turn RP experiments and research into SDFT and self-training for RP behavior shaping.
Important Considerations
As an experimental release, outputs may still exhibit instability, repetition, or stylistic unevenness. The underlying Heretic SDFT base intentionally has weakened guardrails, and users are responsible for the content generated. It is provided for research and evaluation purposes only.