Hugging Face Overview
Overview
NVIDIA's OpenMath-Nemotron-14B-Kaggle is a 14.8 billion parameter model built upon the Qwen2.5-14B architecture. It was fine-tuned using a subset of the OpenMathReasoning dataset, specifically designed to excel in mathematical reasoning tasks. This model played a key role in NVIDIA's first-place achievement in the AIMO-2 Kaggle competition.
Key Capabilities
- Advanced Mathematical Reasoning: Optimized for solving complex math problems, as evidenced by its performance in the AIMO-2 Kaggle competition.
- Benchmark Performance: Achieves competitive results on mathematical benchmarks such as AIME24, AIME25, HMMT-24-25, and HLE-Math, often outperforming other models in its class.
- High Context Length: Supports a context length of up to 131,072 tokens, allowing for processing extensive mathematical problems and related information.
- Code Execution Integration: Designed to work effectively with code execution for solving math problems, with a reference implementation available in NeMo-Skills.
Good For
- Mathematical Research: Intended to facilitate research and development in the field of mathematical reasoning.
- Competitive Problem Solving: Highly suitable for tasks requiring robust mathematical problem-solving capabilities, such as those found in mathematical olympiads.
- Specialized Math Applications: Best utilized in scenarios where the primary requirement is accurate and deep mathematical understanding and solution generation, rather than general conversational AI.