DarwinAnim8or/Trouper-12B
DarwinAnim8or/Trouper-12B is a 12 billion parameter character roleplay model, fine-tuned from Mistral-Nemo-Base-12B on a custom "Actors" dataset. It is specifically designed to produce natural, human-like prose with minimal AI-typical patterns, excelling in character depth, emotional range, and authentic dialogue. This model is optimized for efficient inference in character roleplay scenarios, particularly when paired with the Mistral-V3-Tekken template.
Loading preview...
Overview
DarwinAnim8or/Trouper-12B is a 12 billion parameter character roleplay model, fine-tuned from Mistral-Nemo-Base-12B. It was developed to improve upon previous models like TinyRP-12B, focusing on overcoming limitations and utilizing a new, custom "Actors" dataset. A key design choice was training on a base model to avoid "GPT-isms" and ensure a more malleable foundation for character-specific behavior.
Key Capabilities
- Natural Prose: Generates direct, concrete descriptions and avoids formulaic AI writing patterns, resulting in more human-like text.
- Character Depth: Excels at handling emotional progression, vulnerability, and character growth throughout conversations.
- Authentic Dialogue: Produces natural-sounding speech patterns that are not performative.
- Efficient Roleplay: At 12B parameters, it offers faster inference compared to larger models while maintaining high-quality character interactions.
- Structural Variety: Avoids repetitive response patterns, keeping conversations dynamic.
Differentiators & Use Cases
Trouper-12B is specifically engineered for character roleplay, prioritizing prose quality and emotional depth. It is less prone to "AI slop" compared to its larger counterpart, Prima-24B, due to less synthetic data exposure. While it requires careful template setup (Mistral-V3-Tekken or ChatML are recommended), this model shines in scenarios demanding nuanced emotional roleplay and natural dialogue over sheer reliability or action-oriented RP. It effectively handles 15-20+ turn conversations without character stagnation or formulaic responses, a significant improvement over TinyRP-12B. This model is not intended for general assistant tasks, but rather for dedicated character interaction.