benjamin/Gemma2-2B-Distilled-Math
benjamin/Gemma2-2B-Distilled-Math is a 2.6 billion parameter language model, distilled from OpenMath2-Llama3.1-8B and based on the Gemma2-2B architecture. This model is specifically optimized for mathematical tasks, demonstrating significantly enhanced performance on benchmarks like GSM8K and MATH compared to general-purpose Gemma2 instruction-tuned models. It is designed for applications requiring robust mathematical problem-solving capabilities.
Loading preview...
Model Overview
benjamin/Gemma2-2B-Distilled-Math is a 2.6 billion parameter model derived from the Gemma2-2B base architecture. It has been specifically distilled from the more powerful OpenMath2-Llama3.1-8B model, leveraging the OpenMathInstruct-2 dataset for its training. This distillation process aims to create a smaller, more efficient model that retains strong mathematical reasoning abilities.
Key Capabilities and Performance
This model is engineered to excel in mathematical problem-solving. Its primary differentiator is its specialized performance on math benchmarks, where it significantly outperforms the original general-purpose Gemma2-2B instruction-tuned variant. Key performance metrics include:
- GSM8K (zero-shot): Achieves 65.1%, a substantial improvement over the original Gemma2-2B-IT's 6.1%.
- MATH (zero-shot): Scores 52.1%, compared to 9.9% for the original Gemma2-2B-IT.
Use Cases
This model is particularly well-suited for applications requiring accurate and efficient mathematical problem-solving. Developers should consider this model for tasks such as:
- Automated math problem solvers.
- Educational tools that require step-by-step mathematical reasoning.
- Integration into systems where mathematical understanding is critical, especially in resource-constrained environments where a 2.6B parameter model is advantageous.