nothingiisreal/MN-12B-Celeste-V1.9

5.0 based on 3 reviews
Warm
Public
12B
FP8
32768
Jul 31, 2024
License: apache-2.0
Hugging Face
Overview

Overview

Celeste V1.9 is a 12 billion parameter model developed by nothingiisreal, built upon the Mistral NeMo 12B Instruct architecture. It is specifically fine-tuned for enhanced story writing and roleplaying, leveraging an 8K context window. The training data includes a curated mix of Reddit Writing Prompts (both SFW and NSFW), Kalo's Opus 25K Instruct, and cleaned c2 logs, with a focus on filtering for appropriate message lengths.

Key Capabilities

  • Advanced Story Writing: Excels at generating creative and human-like prose for short stories, with specific system prompts to guide SFW or NSFW content.
  • Dynamic Roleplaying: Designed for engaging roleplay, offering improved NSFW content generation and more active narration compared to previous versions.
  • ChatML Support: Trained with ChatML tokens to prevent EOS bleeding and ensure consistent formatting.
  • Steerability: Highly steerable through OOC (Out of Character) prompts, allowing users to guide character behavior and narrative direction effectively.
  • Flexible Output: Capable of producing varied formatting and message lengths, adapting to initial user input.

Good For

  • Creative Writers: Ideal for generating story continuations, character dialogues, and descriptive narratives.
  • Roleplay Enthusiasts: Provides a robust platform for immersive character interactions and scenario development.
  • Users Seeking NSFW Content: Offers improved handling of NSFW themes while maintaining narrative quality, with options for slow-burn or explicit content based on user preference.
  • Experimentation with Prompts: Encourages users to experiment with sampler settings and system prompts to tailor the model's output to specific needs.