Darkest-muse-v1
| Property | Value |
|---|---|
| Base Model | Gemma-2-9b-it |
| Training Data | Gutenberg3 Dataset |
| Model URL | https://huggingface.co/sam-paech/Darkest-muse-v1 |
| GGUF Version | Available at mradermacher/Darkest-muse-v1-GGUF |
What is Darkest-muse-v1?
Darkest-muse-v1 is a creative writing AI model that combines two distinct personalities: the natural, spare prose style of Quill-v1 and the dark, experimental flair of Delirium-v1. Fine-tuned on the Gutenberg3 dataset containing 629 public domain fiction novels, this model creates unique narrative voices while avoiding common AI writing pitfalls like vocabulary flexing and excessive positivity bias.
Implementation Details
The model is implemented as a merge using the dare_ties method, combining Quill-v1 (density: 0.53, weight: 0.5) and Delirium-v1 (density: 0.53, weight: 0.4) with Ataraxy-v2 as a foundation. It uses bfloat16 dtype and implements int8_mask for optimization.
- Removes common AI writing artifacts and biases
- Balanced merge of natural prose and experimental writing
- Implements Gemma instruction template
- Scored highly on EQ-Bench creative writing tests
Core Capabilities
- Natural, human-like prose generation
- Dark and experimental narrative styles
- Strong dialogue writing abilities
- Stream of consciousness writing
- Reduced vocabulary flexing and artificial positivity
Frequently Asked Questions
Q: What makes this model unique?
The model's unique strength lies in its ability to blend natural prose with experimental writing, creating distinctive narrative voices while avoiding common AI writing patterns. It inherits human-like qualities from the Gutenberg3 dataset while maintaining creative flexibility.
Q: What are the recommended use cases?
The model excels in creative writing tasks, particularly in generating dialogue-rich scenes, historical fiction, and experimental narrative styles. It's particularly suited for writers looking to create unique voices and avoid typical AI writing patterns.
