cgato/L3-TheSpice-8b-v0.8.3

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:8kPublished:Apr 23, 2024License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Warm

cgato/L3-TheSpice-8b-v0.8.3 is an 8 billion parameter Llama 3-based language model developed by cgato, fine-tuned for enhanced roleplay and interactive narrative experiences. It features a unique ability to provide detailed narration, character thoughts, and summaries upon request, making it distinct from general-purpose LLMs. This model is optimized for flexible and immersive conversational interactions, particularly in creative and story-driven applications.

Loading preview...

Overview

cgato/L3-TheSpice-8b-v0.8.3 is a Llama 3-based model, fine-tuned with a focus on creating a more flexible and unique interactive experience, particularly for roleplay and narrative generation. The model has undergone a tokenizer fix to align with the base Llama 3 and was trained for three epochs on a curated, smaller dataset emphasizing a "less is more" approach.

Key Capabilities

  • Interactive Narration: The model can narrate details about objects or characters in a scene without necessarily advancing the story, responding to queries like "What do I see?"
  • Character Insight: Users can request to know a character's thoughts or plans, providing deeper immersion into the narrative.
  • Character Summaries: The model can provide quick summaries of characters, allowing for better context before continuing a conversation.
  • Flexible Interaction: Designed to integrate seamlessly into conversational flows, allowing users to interject with specific queries about the scene or characters.

Training Data

The model was trained using a refined dataset that includes Capybara, Claude Multiround 30k, Augmental, ToxicQA, Yahoo Answers, Airoboros 3.1, and LimaRP, with a particular emphasis on LimaRP for its unique characteristics.

Recommended Use Cases

This model is particularly well-suited for applications requiring rich, interactive storytelling, role-playing, and detailed narrative generation where users need to frequently query the environment or character states. It is configured for chat-based prompt formats, compatible with tools like Oobabooga and Silly Tavern.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p