Dionysus-Mistral-m3-v6: A Roleplay-Optimized Language Model
Dionysus-Mistral-m3-v6 is a 7 billion parameter language model developed by DECEM, built upon the AIDC-ai-business/Marcoroni-7B-v3 base model. This model has undergone Supervised Fine-Tuning (SFT) specifically to excel in roleplay scenarios, making it a specialized tool for interactive narrative generation and character simulation.
Key Capabilities & Performance
- Roleplay Optimization: The primary focus of this model is to provide high-quality responses suitable for roleplaying, indicating a fine-tuning process geared towards conversational depth and character consistency.
- Base Architecture: Leverages the Marcoroni-7B-v3 architecture, suggesting a foundation for robust language understanding and generation.
- Evaluation Metrics: On the Open LLM Leaderboard, Dionysus-Mistral-m3-v6 achieved an average score of 67.10. Notable individual scores include 63.14 on AI2 Reasoning Challenge (25-Shot), 84.51 on HellaSwag (10-Shot), and 62.82 on MMLU (5-Shot), indicating solid performance across various benchmarks for a model of its size.
Prompting Guidelines
The model is designed to be used with an Alpaca-style prompt template, facilitating clear instruction and response formatting for roleplay interactions. Users should structure prompts with a system prompt, followed by ### Instruction: {user_name}: {prompt}, and expect responses formatted as ### Response: {character_name}:.
Ideal Use Cases
- Interactive Storytelling: Generating dynamic and engaging narratives where users interact with AI-controlled characters.
- Character Simulation: Creating AI personas that maintain consistent traits, dialogue styles, and backstories for immersive experiences.
- Creative Writing Assistance: Aiding writers in developing character dialogue and plot points within a roleplay context.