nbeerbower/Vitus-Qwen3-14B
Vitus-Qwen3-14B by nbeerbower is a 14 billion parameter Qwen3-based language model, fine-tuned using ORPO on the human-writing-dpo dataset. This model is specifically optimized for generating high-quality, creative long-form text, demonstrating strong narrative quality and prose style. It excels in imaginative writing tasks, particularly when generating detailed and emotionally resonant stories.
Loading preview...
Vitus-Qwen3-14B: Optimized for Creative Long-Form Writing
Vitus-Qwen3-14B is a 14 billion parameter model developed by nbeerbower, built upon the Qwen3 architecture. This model has been specifically fine-tuned using the ORPO (Optimized Reward-Prompt Optimization) method over two epochs on a single RTX A6000 GPU. The training utilized the nbeerbower/human-writing-dpo dataset, indicating a strong focus on enhancing human-like writing capabilities.
Key Capabilities
- Exceptional Narrative Quality: Demonstrates confident pacing, seamless scene-to-scene flow, and strong structural integrity in generated text.
- Lush Prose Style: Produces lyrical and emotionally dense prose with good rhythm and sentence balance, suitable for evocative storytelling.
- Deep Thematic Integration: Capable of weaving complex themes like memory, grief, and duty elegantly into narratives.
- High Prompt Relevance: Excels at adhering to detailed and imaginative prompt instructions, incorporating core concepts with gravitas.
- Speculative Imagination: Shows strong ability in creating unique and potent conceptual frameworks, such as 'storm-as-absence' or 'archive as emotional crypt'.
Evaluation Highlights
The model was evaluated using OpenAI o3 as a judge on a complex creative writing prompt, achieving a total score of 46 out of 50. This high score reflects its proficiency across narrative quality, prose style, thematic depth, prompt relevance, and speculative imagination.
Recommended Use
- Creative Writing: Ideal for generating long-form stories, descriptive passages, and imaginative narratives.
- Content Generation: Suitable for tasks requiring high-quality, engaging, and emotionally resonant text.
For optimal writing results, it is recommended to set enable_thinking to False when using this model.