SicariusSicariiStuff/Dusk_Rainbow
Dusk_Rainbow is an 8 billion parameter language model developed by SicariusSicariiStuff, fine-tuned from Invisietch's EtherealRainbow-v0.3-8B. This model is optimized for creative writing and general tasks, demonstrating exceptional adherence to instructions regarding paragraph length and story writing prompts. It is notable for its very low censorship level (9.1/10) and was trained using a fraction of the data intended for the LLAMA-3_8B_Unaligned project, specifically avoiding API model data to minimize bias and repetitive language.
Loading preview...
Dusk_Rainbow: A Creative Writing and General Task LLM
Dusk_Rainbow is an 8 billion parameter language model developed by SicariusSicariiStuff, built upon Invisietch's EtherealRainbow-v0.3-8B. This model is distinguished by its focus on creative writing and general tasks, showcasing a very low censorship level (9.1/10).
Key Capabilities
- Creative Writing: Excels in generating stories and adhering to specific paragraph length instructions.
- Low Censorship: Designed with minimal content restrictions, offering greater flexibility for diverse outputs.
- Instruction Adherence: Demonstrates strong ability to follow detailed instructions, particularly for structured text generation.
- Unique Training Data: Fine-tuned using a fraction of the data for the LLAMA-3_8B_Unaligned project, specifically avoiding API-generated content to reduce bias and repetitive phrasing.
Good For
- Story Generation: Ideal for users seeking models that can produce imaginative and structured narratives.
- Uncensored Content: Suitable for applications requiring less restrictive content generation.
- Paragraph Control: Useful for tasks where precise control over output paragraph length is crucial.
Benchmarks
The model achieves an average score of 18.49 across various benchmarks, including 35.88 on IFEval (0-Shot) and 25.96 on BBH (3-Shot), indicating its general task performance.