Hugging Face Model Overview
dreamgen/opus-v1.2-7b is a 7 billion parameter model from the DreamGen Opus V1 series, primarily designed for steerable story-writing and role-playing. It employs an extended ChatML format, introducing a text role with optional names to manage character-specific narration and dialogue within a story.
Key Capabilities
- Steerable Story-writing & Role-playing: Users provide plot, style, and character descriptions, then guide the narrative with instructions for story or role-play continuations.
- Story Plot Summarization: Generates concise summaries of stories or chapters, with examples demonstrating 50-word and 200-word outputs.
- Story Character Description: Extracts and describes characters from provided story segments.
- Story Style Description: Analyzes and describes the writing style of a given story.
- Story Description to Chapters: Converts a brief plot description into chapter-by-chapter outlines.
Training and Usage
The model was fine-tuned on approximately 100 million tokens, with examples up to 31,000 tokens long, focusing on human-written story-writing and role-playing data. Recommended sampling parameters for creative generation include min_p in the range [0.01, 0.1] and temperature in [0.5, 1.5]. The model is compatible with various backends, but emphasizes correct prompt formatting and tokenization, particularly for the extended ChatML structure. Resources like a prompting guide and a Google Colab are available for developers.