muverqqw/Noir
muverqqw/Noir is a 3.1 billion parameter Qwen 2.5 architecture model developed by IceL1ghtning, designed as a high-efficiency powerhouse within the Noir series. This model excels in logic and mathematical reasoning, achieving 65.0% on GSM8K, a score typically seen in larger models. It also demonstrates strong creative writing capabilities with a 79.7% narrative evaluation score and stable logical processing, making it suitable for complex multi-step instructions. Noir is optimized for local deployment, fitting into 8GB of VRAM while providing robust performance for technical tasks and creative applications.
Loading preview...
Noir-Standard (3B): The Versatile Workhorse
Noir-Standard, developed by IceL1ghtning, is a 3.1 billion parameter model built on the Qwen 2.5 architecture. It is positioned as the "high-efficiency powerhouse" of the Noir series, designed to deliver strong performance in logic and mathematical reasoning while being optimized for local deployment with 8GB VRAM.
Key Capabilities & Performance:
- Mathematical Reasoning: Achieves an impressive 65.0% on GSM8K, outperforming many 7B and 8B parameter models.
- Creative Writing: Scores 79.7% on Narrative Eval, demonstrating exceptional prose and vocabulary diversity for vivid storytelling.
- Logical Coherence: Maintains stable logic with a 26.0% on ARC (Challenge), enabling it to follow complex multi-step instructions.
- Balanced Knowledge: Provides balanced performance across STEM and Humanities with a 21.0% on MMLU.
Ideal Use Cases:
- Technical Tasks: Excellent for logical problem-solving and mathematical applications.
- Creative Content Generation: Capable of crafting atmospheric stories with distinct character voices.
- Local Deployment: Optimized for fast, lean, and smart operation on systems with limited VRAM.
- "Smart Professional" Assistant: Designed for users needing a capable assistant that fits within 8GB of VRAM while maintaining deep creative flow.