SuperQAI2050/Math
SuperQAI2050/Math is a 32.8 billion parameter language model developed by NVIDIA, fine-tuned from Qwen2.5-32B on the OpenMathReasoning dataset. This model is specifically optimized for advanced mathematical reasoning tasks, achieving state-of-the-art results on benchmarks like AIME24 and HMMT-24-25. It supports a 32768-token context length and is designed for commercial use in mathematical problem-solving.
Loading preview...
OpenMath-Nemotron-32B: Advanced Mathematical Reasoning
SuperQAI2050/Math, also known as OpenMath-Nemotron-32B, is a 32.8 billion parameter model developed by NVIDIA, built upon the Qwen2.5-32B architecture. It has been extensively fine-tuned using the OpenMathReasoning dataset to excel in complex mathematical problem-solving.
Key Capabilities & Performance
This model demonstrates state-of-the-art performance on challenging mathematical benchmarks, including AIME24, AIME25, HMMT-24-25, and HLE-Math. It supports various inference modes:
- Chain-of-Thought (CoT): For step-by-step reasoning.
- Tool-Integrated Reasoning (TIR): Integrating external tools for enhanced problem-solving.
- Generative Solution Selection (GenSelect): A method that significantly boosts accuracy, with the 32B GenSelect variant achieving up to 93.3% on AIME24 and 73.5% on HMMT-24-25.
Notably, a version of the 14B model secured first place in the AIMO-2 Kaggle competition. The entire pipeline, including code, models, and dataset, is open-sourced for reproducibility.
Use Cases & Limitations
This model is primarily intended for research and applications requiring advanced mathematical reasoning. It is ready for commercial use. However, it's important to note that it has not been instruction-tuned on general data and may not perform optimally outside the mathematical domain. The model supports a context length of up to 32768 tokens.