Dusk_Rainbow
| Property | Value |
|---|---|
| Parameter Count | 8.03B |
| Model Type | Language Model |
| License | Apache 2.0 |
| Tensor Type | FP16 |
What is Dusk_Rainbow?
Dusk_Rainbow is a specialized language model built on the LLaMA architecture, fine-tuned with 16M tokens of creative writing data. It's designed for uncensored creative writing and general tasks, featuring exceptional instruction following capabilities and natural storytelling abilities.
Implementation Details
The model is based on EtherealRainbow-v0.3-8B and has been further refined with carefully curated training data. Unlike many other models, it avoids using ChatGPT or Claude-generated content in its training data to maintain originality and reduce repetitive patterns.
- Available in multiple formats including GGUF, EXL2, and specialized FP8 versions
- Optimized for both desktop and mobile (ARM) deployments
- Supports multiple instruction templates including ChatML and Llama-3
Core Capabilities
- Creative writing with strong paragraph structure adherence
- Story continuation and development
- Poetry generation with analytical capabilities
- Low censorship level (9.1/10 on the uncensored scale)
- Strong performance on various benchmarks including IFEval and BBH
Frequently Asked Questions
Q: What makes this model unique?
The model stands out for its ability to maintain consistent paragraph structure and creative writing capabilities without relying on API-generated training data, resulting in more original and less repetitive outputs.
Q: What are the recommended use cases?
Dusk_Rainbow excels in creative writing tasks, story generation, and general instruction following. It's particularly well-suited for users who need detailed, structured creative content with minimal censorship.
