logihertz/nyra-A
Nyra-A is an 8 billion parameter reasoning model developed by Logihertz Systems, based on an optimized Llama-3 architecture. It is specifically engineered as a "Primary Logic Core" for high-precision tasks, excelling in mathematical consistency, structured data processing, and complex logical deduction. This model is optimized for generating precise JSON, XML, and code structures, and for analytical processing where hallucination must be minimized. With a 32768 token context length, it is ideal for applications requiring rigorous algorithmic reasoning and structured output.
Loading preview...
Nyra-A: The Logic Core
Nyra-A is an 8 billion parameter reasoning model developed by Logihertz Systems as part of the independent Nyra Project. It is built on an optimized Llama-3 architecture, utilizing DARE-TIES + SLERP merge methodology for enhanced weight-sum stability. This model is designed as a "Primary Logic Core" (Tier A), focusing on high-performance logical deduction and mathematical consistency.
Key Capabilities
- Algorithmic Reasoning: Solves complex mathematical and logical proofs with high precision.
- Structured Output: Generates accurate JSON, XML, and intricate code structures.
- Analytical Processing: Refines complex multi-turn instructions, minimizing hallucination.
Intended Use Cases
Nyra-A is best suited for standalone applications demanding high precision and reliability in outputs. While currently undergoing rigorous evaluation with benchmarks like MMLU-Pro, GSM8K, and HumanEval, its core strength lies in tasks requiring strict logical adherence and structured data handling. Users should implement secondary validation for critical deployments due to potential inherited biases from its foundational weights.