Heralax/MythoMakiseMerged-13b
Heralax/MythoMakiseMerged-13b is a 13 billion parameter language model, a fine-tuned variant of MythoMax-L2-13b, optimized for engaging conversation and roleplay. It was trained on a GPT-4 processed visual novel script, resulting in a model that excels at banter and character roleplaying. This version incorporates a 33% merge back with MythoMax-L2-13b to enhance intelligence and stability while retaining its distinct personality and conversational style.
Loading preview...
MythoMakiseMerged-13b Overview
Heralax/MythoMakiseMerged-13b is a 13 billion parameter model derived from MythoMax-L2-13b, specifically fine-tuned for high-quality conversational and roleplaying interactions. The model's training involved approximately 1300 high-quality examples generated by GPT-4 from a visual novel script, focusing on grouping conversations into effective training examples.
Key Capabilities & Features
- Exceptional Roleplay: Despite initial training for a specific character, the model demonstrates strong capabilities in roleplaying various characters, particularly those with similar personality traits to its original target.
- Engaging Conversation: Excels in banter and general conversational tasks, making it suitable for interactive applications.
- Uncensored Output: The model is designed to be uncensored, offering flexibility in content generation.
- Enhanced Stability & Intelligence: This merged version incorporates 33% of the original MythoMax-L2-13b, improving overall intelligence and stability while preserving its unique personality and style.
- SillyTavern Prompt Format: Optimized for the SillyTavern prompt format, ensuring better performance within that frontend.
Good For
- Roleplaying Scenarios: Ideal for generating dynamic and engaging character interactions.
- Conversational Agents: Suitable for chatbots or applications requiring natural and bantering dialogue.
- Creative Writing Assistance: Can be used to generate character dialogue or narrative elements where personality and conversational flow are key.
While proficient in conversation and roleplay, users should not expect high accuracy in factual recall or other non-conversational tasks.