TheSpice-7b-v0.1.1: A Narrative-Focused LLM
TheSpice-7b-v0.1.1, developed by cgato, is a 7 billion parameter language model with a 4096-token context length, specifically designed to offer a flexible and unique interactive narrative experience. This model represents a return to a "less is more" training approach, utilizing a small, hand-edited dataset that includes Dolphin, Ultrachat, Capybara, Augmental, ToxicQA, Yahoo Answers, and Airoboros 3.1.
Key Capabilities
- Dynamic Narration: The model can generate detailed narration about objects or characters in a scene without necessarily advancing the plot, responding to queries like "What do I see?"
- Character Insight: Users can inquire about a character's thoughts or plans, allowing for deeper engagement with the narrative.
- Character Summaries: The model can provide quick summaries of characters, enhancing roleplay and story understanding.
- Flexible Conversation Flow: These narrative and insight features are integrated into a continuous conversational flow, allowing users to seamlessly transition between inquiry and story progression.
Recommended Usage
This model is optimized for chat and roleplay scenarios, particularly within platforms like Oobabooga and SillyTavern. The developer provides recommended preset settings for these platforms (e.g., Temp: 1.25, MinP: 0.1, RepPen: 1.05 for SillyTavern) to achieve the intended interactive experience. The prompt format follows a standard chat template, expecting Username: {Input}\nBotName: {Response}.