Casual-Autopsy/L3-Umbral-Mind-RP-v2.0-8B: A Specialized Role-Play Model
This 8 billion parameter model, created by Casual-Autopsy, is a merge of several pre-trained language models, specifically engineered for nuanced role-playing experiences. Its primary differentiator is its ability to handle heavy and sensitive themes like mental illness, self-harm, trauma, and suicide, addressing a common limitation in other RP models that tend to be overly positive.
Key Capabilities
- Specialized Role-Play: Designed for role-plays involving complex and dark themes, offering a more realistic and less saccharine interaction.
- Savior/Reverse Savior Scenarios: Particularly suited for narratives where characters engage in savior or reverse savior dynamics.
- Format Compatibility: Optimized for use with asterisk/quotes role-playing formats.
- Merge Architecture: Built using multiple Task Arithmetic and Model Stock merges, incorporating models with psychology data to enhance its thematic understanding.
Evaluation Highlights
On the Open LLM Leaderboard, the model achieved an IFEval (0-Shot) score of 71.23, which is highlighted as the most crucial metric for RP AIs, indicating its proficiency in following OOC, Lorebooks, and character cards. Other metrics like BBH (3-Shot) at 32.49 and MMLU-PRO (5-shot) at 30.26 are also reported.
When to Use This Model
This model is ideal for users seeking an RP partner capable of engaging deeply and realistically with challenging emotional and psychological themes, particularly within savior/reverse savior narratives, and who prefer asterisk/quotes RPing styles.