digitous/Alpacino13b
The digitous/Alpacino13b is a 13 billion parameter language model based on the Llama architecture, created by digitous. It is a triple merge of Alpaca, CoT (Chain-of-Thought), and Storytelling LoRAs, designed to enhance reasoning and story writing capabilities while maintaining Alpaca's instruct format. This model is primarily optimized for generating verbose, detailed, and creative narrative responses, making it suitable for text-based adventure games and creative writing applications.
Loading preview...
Alpacino13b: Enhanced Narrative and Reasoning
The digitous/Alpacino13b is a 13 billion parameter model built upon the Llama architecture, developed by digitous. Its core innovation lies in a triple merge of LoRAs: Alpaca, Chain-of-Thought (CoT), and Storytelling. This unique combination aims to significantly boost both the reasoning and story writing abilities of the base Alpaca model, while preserving its instruction-following format.
Key Capabilities
- Enhanced Reasoning: Integrates Chain-of-Thought capabilities to improve logical processing.
- Advanced Storytelling: Optimized for generating verbose, detailed, and creative narrative content.
- Instruction Following: Retains the strong instruct-following characteristics of the Alpaca base model.
- Text-Based Adventure Generation: Specifically designed to function as a narrator for interactive text-based adventure games, producing rich and imaginative descriptions.
Good For
- Creative Writing: Ideal for generating long-form creative content, stories, and descriptive passages.
- Interactive Narratives: Excels in applications requiring dynamic and detailed responses for text-based adventure games or interactive fiction.
- Instruction-Tuned Tasks: Benefits from the Alpaca backbone for general instruction-following tasks, with added narrative flair.
Usage Notes
For optimal performance in text-based adventure scenarios, the model suggests using "Storywriter" or "Godlike" presets in Text-Generation-WebUI or KoboldAI, with context tokens around 2048 and max generation tokens at approximately 680 or greater. The model is under a non-commercial license and requires users to have explicit access to the original Llama weights from Meta AI.