NeuR0mancR/Neural-v1-24B
Neural-v1-24B by NeuR0mancR is a 24 billion parameter Mistral Small 3.2 merge model, specifically designed for narrative generation and roleplay with a 32768 token context length. It specializes in producing atmospheric, world-weary, and analytical prose, prioritizing narrative realism over general-purpose assistance. This model is an unaligned research artifact, lacking safety guardrails and focusing on stylistic "grit" for creative and research applications.
Loading preview...
Overview
NeuR0mancR/Neural-v1-24B is a 24 billion parameter language model based on the Mistral Small 3.2 architecture, developed by NeuR0mancR. It is a specialized merge model, not intended as a general-purpose assistant, but rather fine-tuned for narrative generation and roleplay. The model excels at producing prose with an atmospheric, world-weary, and analytical style, prioritizing narrative realism.
Key Capabilities & Characteristics
- Specialized Narrative Generation: Focuses on creating immersive and stylistically distinct narratives.
- Roleplay Optimization: Designed to enhance roleplaying scenarios with its unique prose style.
- Unaligned Research Artifact: Functions as a next-token predictor without safety alignment or standard guardrails, allowing for exploration of themes like violence, horror, sexuality, and dark psychological tension.
- Technical Foundation: Built upon Mistral Small 3.2 (24B) using a multi-merge strategy (Model Stock, DARE Ties, nuSLERP & SLERP) to preserve knowledge and minimize "slop."
- Context Length: Supports a 32768 token context window.
Intended Use Cases
- Creative Writing: Ideal for authors and writers seeking a model that can generate atmospheric and gritty prose.
- Roleplay Scenarios: Suitable for developers and users who require a model that prioritizes narrative depth and realism in roleplaying.
- Research & Experimentation: Serves as a base model for further stylistic fine-tuning and research into unaligned language model behaviors.
Note: Due to its unaligned nature, users deploying this model in public-facing applications are advised to implement their own moderation and safety measures.