darthcrawl/artifex-rp-orpheus-llama-3.1-8b
darthcrawl/artifex-rp-orpheus-llama-3.1-8b is an 8 billion parameter Llama-3.1-8B-Instruct fine-tune by darthcrawl, specifically optimized for generating vivid, intelligent, and restrained literary prose. This model excels at crafting intimate fiction with concrete sensory detail, rhythmic flow, and nuanced tension, making it a specialized writing partner rather than a general chat assistant. It is part of the Artifex RP series, designed for common hardware and focused on specific narrative voices.
Loading preview...
Model Overview
darthcrawl/artifex-rp-orpheus-llama-3.1-8b is a specialized fine-tune of the meta-llama/Llama-3.1-8B-Instruct base model, developed by darthcrawl. This 8 billion parameter model is part of the Artifex RP series, which focuses on creating distinct narrative voices optimized for various hardware tiers. Orpheus is specifically designed to produce restrained-lyrical prose, emphasizing vivid, intelligent writing where tension is paramount.
Key Capabilities
- Literary Prose Generation: Shifts the model's output towards intimate fiction with a focus on nuanced expression.
- Tension and Subtlety: Excels at creating narrative tension, implying meaning rather than explicitly stating it.
- Sensory Detail: Generates concrete sensory descriptions, grounding the narrative in tangible experiences.
- Rhythmic Writing: Produces prose with varied sentence structures and natural rhythm.
- Broad Range: Capable of handling diverse fictional scenarios, from quiet domestic scenes to charged romantic moments and physical interactions, all within its defined register.
Training Details
The model was fine-tuned using LoRA SFT on a single H100 GPU. Its corpus consisted of approximately 1,600 synthesized passages across five register categories, including non-romantic, romantic-charged, physical-restrained, quiet-quotidian, and vocabulary-uplift content, generated by advanced LLMs like Claude Sonnet 4.6, Claude Opus 4.6, and DeepSeek V3, alongside handwritten explicit scenes for range.
Good For
- Creative Writing: Ideal for authors and developers seeking a writing partner to generate high-quality, literary-style fiction.
- Roleplay Integration: Designed to apply its unique prose register to character cards and system prompts in roleplaying environments like SillyTavern, Kobold, or Ollama.
- Specific Narrative Voice: Users needing a model that produces intelligent, warm, and restrained prose without lecturing or purple prose.
Limitations
Due to its small training corpus, the model may exhibit drift on very long generations. It is optimized for third-person literary prose, with first-person and chat-style outputs being less tuned. It is not intended as a factual assistant, and its reasoning, knowledge, math, and coding capabilities are inherited from the base Llama 3.1 model, neither improved nor degraded.