nothingiisreal/Celeste-12B-V1.6

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:12BQuant:FP8Ctx Length:32kPublished:Jul 28, 2024License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

Celeste-12B-V1.6 by nothingiisreal is a 12 billion parameter instruction-tuned model based on Mistral NeMo 12B Instruct, fine-tuned for creative writing and roleplay with an 8K context window. This version emphasizes high coherency and strong adherence to 'OOC:' instructions, making it highly steerable for narrative generation. It excels at generating diverse prose and adapting to various writing styles, particularly for interactive storytelling and character-driven scenarios.

Loading preview...

nothingiisreal/Celeste-12B-V1.6: A Roleplay and Creative Writing Powerhouse

Celeste-12B-V1.6 is a 12 billion parameter instruction-tuned model built upon the Mistral NeMo 12B Instruct architecture, developed by nothingiisreal. It has been specifically fine-tuned for creative writing and roleplaying, leveraging an 8K context window. This iteration, V1.6, is noted for its significantly improved coherency and exceptional ability to follow OOC: (Out of Character) instructions, allowing for precise steering of the model's output.

Key Capabilities & Features

  • High Coherency: Delivers consistent and logical narratives, even in complex roleplay scenarios.
  • Exceptional OOC Steering: Responds very well to OOC: prompts for guiding character behavior or narrative direction.
  • Creative Writing & Roleplay: Optimized for generating diverse prose, adapting to various writing styles, and maintaining character personas.
  • Flexible Content Generation: Capable of handling both SFW and NSFW content, with specific tips provided for controlling 'horniness' and refusals.
  • Context Length: Trained on 8K context, with potential to inherit longer context from its L3.1 base.
  • Training Data: Fine-tuned on a curated mix of Reddit Writing Prompts, Kalo's Opus 25K Instruct, and c2 logs cleaned, with specific filtering to optimize for 8B model performance.

When to Use This Model

Celeste-12B-V1.6 is ideal for developers and users focused on:

  • Interactive Storytelling: Creating dynamic and engaging roleplay experiences.
  • Character-Driven Narratives: Generating responses that deeply embody character personas and evolve with the story.
  • Creative Content Generation: Producing varied and imaginative text, from short stories to complex dialogues.
  • Highly Steerable AI: When precise control over the model's output and behavior is crucial, especially through OOC: commands.

Users are encouraged to utilize the provided usage tips, including specific sampler settings and a recommended system prompt, to maximize the model's performance and achieve the desired output quality.

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