| Property | Value |
|---|---|
| Base Model | Small-3.1 Instruct |
| Parameters | 24B |
| Author | Gryphe |
| Model URL | huggingface.co/Gryphe/Pantheon-RP-1.8-24b-Small-3.1 |
| Format | ChatML |
What is Pantheon-RP-1.8-24b-Small-3.1?
Pantheon-RP is a specialized roleplay model designed to provide immersive character interactions through pre-defined personas. Built on the Small-3.1 Instruct base, it combines synthetic data from multiple sources including Sonnet 3.5 + 3.7, ChatGPT 4o, and Deepseek, processed through an extensive rewriting pipeline to eliminate common AI clichés.
Implementation Details
The model implements a ChatML format and features recommended inference settings including temperature of 0.8, repetition penalty of 1.05, and min_p of 0.05. It supports both markdown and novel-style formatting, with a 30/70 ratio favoring novel-style output.
- Multiple persona system with simple activation phrases
- Comprehensive character templates including 12 distinct personas
- Support for both second-person and traditional narrative styles
- Enhanced contextual understanding through detailed prompt templates
Core Capabilities
- Dynamic character roleplaying with consistent personality traits
- Flexible formatting support for different storytelling styles
- Deep contextual understanding of character backgrounds and settings
- Ability to maintain character consistency across conversations
- Support for detailed character and location descriptions
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its ability to switch between multiple pre-defined personas with consistent personality traits, combined with its extensive training on diverse roleplay data sources and custom rewriting pipeline to avoid common AI stereotypes.
Q: What are the recommended use cases?
The model is specifically designed for immersive roleplay scenarios, character-based storytelling, and interactive narrative experiences. It excels in maintaining consistent character personalities while engaging in dynamic conversations.