Opus V1.2 Llama 3 8B Overview
This model, dreamgen/opus-v1.2-llama-3-8b, is an 8 billion parameter variant based on the Llama 3 architecture, developed by dreamgen. It is primarily designed and fine-tuned for steerable story-writing and role-playing, offering capabilities to generate dynamic and interactive narratives. The model utilizes an extended ChatML format, incorporating a text role with optional names to manage character interactions and narration within a story.
Key Capabilities
- Steerable Story-writing and Role-playing: Generates story continuations and role-play responses based on detailed system prompts (plot, style, characters) and user instructions.
- Story Plot Summarization: Condenses stories or chapters into concise plot descriptions of varying lengths.
- Story Character Description: Extracts and describes specific characters from a given story.
- Story Style Description: Analyzes and describes the writing style of a provided narrative.
- Story Description to Chapters: Breaks down a brief plot description into individual chapter descriptions.
Training and Usage
The model was fine-tuned on approximately 100 million tokens, with examples up to 31,000 tokens long, focusing on steerable story-writing, role-playing, and general writing assistance. All story-writing and role-playing examples were derived from human-written text. For optimal performance in story-writing and role-play, dreamgen recommends using "Min P" based sampling with min_p in the range [0.01, 0.1] and temperature in [0.5, 1.5].
Important Note
Users are advised that this specific model version has known issues. dreamgen recommends using newer preview models built on Llama 3 8B Base or Llama 3 8B Instruct for improved performance and stability. Refer to the official discussion for guidance on resolving potential continuous output generation.