13B-HyperMantis Overview
digitous/13B-HyperMantis is a 13 billion parameter language model created through a weight-sum merge of several other 13B models: MantiCore3E, VicunaCocktail, SuperCOT, StorytellingV2, and BluemoonRP. This unique combination aims to leverage the strengths of its constituent models, particularly excelling in creative and interactive text generation tasks.
Key Capabilities and Characteristics
- Optimized for Roleplay and Narrative Generation: The model demonstrates strong performance in generating detailed scenes, character actions, and immersive narratives, making it suitable for text-based adventure games and roleplaying scenarios.
- Alpaca Instruct Format Preference: Despite its core being derived from uncensored Vicuna models, HyperMantis responds best to the formality of the Alpaca instruct format. This is noted to avoid vestigial traces of moralizing often triggered by Human/Assistant formats, which are not conducive to freeform instructions or roleplay.
- Versatile Use Cases: Beyond roleplay, the model is also suggested for use as a regular chatbot, for creating 'waifu' or 'husbando' characters, and for general freeform instruction-following.
- Context Length: The model supports a context length of 4096 tokens, allowing for moderately long and coherent interactions.
Recommended Usage
HyperMantis is particularly well-suited for applications requiring rich, descriptive storytelling and interactive conversational experiences. Users are advised to utilize the Alpaca instruct format for optimal performance, especially when aiming for creative or roleplay-centric outputs. Specific settings for platforms like KoboldAI are provided in the original documentation to enhance its performance in these areas.