muverqqw/Noir-mini

Noir-Mini: The Balanced Thinker

Noir-Mini, developed by IceL1ghtning, is a 1.5 billion parameter model built on the Qwen 2.5 architecture, designed to be a "Reasoning Assistant." It significantly advances logic and mathematical reasoning compared to smaller models, often explaining its logic before providing an answer.

Key Capabilities

• Exceptional Mathematical Reasoning: Achieves a 54.0% score on GSM8K, outperforming most models in its size class for complex multi-step problems.
• Reasoning-First Approach: Prioritizes explaining its logic, making it robust for real-world applications where understanding the "why" is crucial.
• High Creativity: Boasts a creativity score of 72.3%, ensuring fluid, diverse, and non-repetitive prose.
• Efficient Performance: Small enough to run on devices like phones or 4GB GPUs, yet powerful enough to manage complex system prompts.
• Bilingual Support: Optimized for logic-heavy instructions and supports both English and Russian.

Good For

• Tasks requiring detailed logical explanations and mathematical problem-solving.
• Applications needing a balance of speed and intelligence on resource-constrained devices.
• Generating creative and diverse text outputs without common smaller model repetitions.