Kyro-n1-14B: Reasoning-Optimized Language Model

Kyro-n1-14B is a 14.8 billion parameter causal language model developed by Spestly and Kazex of Open-Neo. It is a fine-tuned variant of Qwen/Qwen2.5-14B-Instruct, with a primary focus on improving reasoning quality, particularly in mathematics and general conversational contexts. The model is designed to be lightweight and fast, aiming to make advanced reasoning capabilities accessible across a wide range of computing devices.

Key Capabilities

• Enhanced Reasoning: Specifically optimized for mathematical and general reasoning tasks.
• Large Context Window: Supports a full input context length of 131,072 tokens, allowing for processing extensive information.
• Qwen2.5 Architecture: Built upon the robust Qwen2.5 framework, incorporating features like RoPE, SwiGLU, and RMSNorm.
• Multilingual Support: Processes and generates multilingual text and code.

Good For

• Applications requiring strong mathematical problem-solving.
• Use cases demanding advanced logical reasoning in conversational AI.
• Deployment on devices with varying compute resources, thanks to its optimized design and availability in multiple sizes (3B, 7B, 14B).