Casual-Autopsy/L3-Deluxe-Scrambled-Eggs-On-Toast-8B
Casual-Autopsy/L3-Deluxe-Scrambled-Eggs-On-Toast-8B is an 8 billion parameter Llama 3-based model merger, created from 36 distinct models over 23 merging steps. This model is specifically designed for role-play scenarios, aiming to balance creativity and intelligence through a gradient-weighted merging approach. It leverages various Llama 3-based models to enhance both imaginative output and adherence to instructions, making it suitable for nuanced conversational and narrative generation.
Loading preview...
Model Overview
Casual-Autopsy/L3-Deluxe-Scrambled-Eggs-On-Toast-8B is an 8 billion parameter model built upon the Llama 3 architecture. It is the result of an extensive merging process, combining 36 different models over 23 distinct steps using LazyMergekit. The core philosophy behind this model is to achieve a harmonious blend of creativity and intelligence, particularly for role-playing applications.
Key Capabilities
- Role-Play Optimization: Specifically engineered for rich and dynamic role-playing interactions.
- Balanced Output: Utilizes a gradient-weighting system during merging to promote both creative expression and intelligent, adherent responses.
- Complex Merging Strategy: Incorporates multiple merging methods (dare_ties, breadcrumbs_ties, slerp) across various stages to integrate diverse model strengths.
- Llama 3 Foundation: Benefits from the underlying capabilities of the Llama 3 base models.
Recommended Use Cases
- Creative Role-Playing: Ideal for generating engaging and imaginative character interactions and narratives.
- Interactive Storytelling: Suitable for applications requiring dynamic and context-aware story generation.
- Conversational AI: Can be used in scenarios where a balance of free-form creativity and logical coherence is desired.
Configuration
The model provides recommended sampler settings for both "more creativity" and "more adherencey," allowing users to fine-tune its behavior based on their specific needs. It is designed to work with the Llama 3 instruct format, with Virt-io's SillyTavern Presets recommended for instruct/context settings.