Casual-Autopsy/L3-Umbral-Mind-RP-v1.0-8B

Model Overview

Casual-Autopsy/L3-Umbral-Mind-RP-v1.0-8B is an 8 billion parameter merged language model, created by Casual-Autopsy using MergeKit. It is specifically engineered for role-playing (RP) applications, particularly those involving heavy and sensitive themes like mental illness, self-harm, trauma, and suicide. The model aims to provide a more realistic and less overtly positive portrayal of these topics, addressing a common limitation in other RP models.

Key Capabilities

• Specialized Role-Playing: Designed to handle complex and dark themes with greater nuance, avoiding overly optimistic or saccharine responses.
• UGI Leaderboard Performance: Achieved the highest ranking among 8B models on the UGI Leaderboard at the time of its release.
• Merge Architecture: Constructed from multiple base models, including failspy/Llama-3-8B-Instruct-MopeyMule and models with psychology data, using a combination of Task Arithmetic and Model Stock merge methods.

Usage and Format

This model is intended for use with asterisks/quotes RPing formats. Using other formats may lead to suboptimal performance or issues. The merge process incorporated a model specifically fine-tuned for this format (ResplendentAI/RP_Format_QuoteAsterisk_Llama3).

Good For

• Users seeking an RP model capable of engaging with dark, heavy, and emotionally challenging themes without sanitization.
• Role-players interested in "savior/reverse savior" type narratives.
• Developers looking for a specialized 8B model for nuanced character interactions in sensitive contexts.