fdyrd/QwenMath-0.5B
QwenMath-0.5B by fdyrd is a 0.5 billion parameter generative language model specifically designed and fine-tuned for solving mathematical problems. Utilizing a Qwen-based architecture, this model demonstrates capabilities across various math domains including algebra, geometry, and number theory. It is optimized for mathematical reasoning tasks, offering a compact solution for applications requiring math problem-solving. The model has a context length of 32768 tokens.
Loading preview...
Overview
fdyrd/QwenMath-0.5B is a 0.5 billion parameter generative language model developed by fdyrd, specifically engineered for mathematical problem-solving. It leverages a Qwen-based architecture and has been fine-tuned using LoRA on a dataset of 500 math problems over 3 epochs. The model is designed to handle a wide range of mathematical challenges, from basic algebra to more complex precalculus and number theory.
Key Capabilities
- Mathematical Problem Solving: Specialized in generating solutions for math problems across multiple domains.
- Domain Coverage: Demonstrates performance in Algebra, Intermediate Algebra, Prealgebra, Precalculus, Number Theory, Geometry, and Counting & Probability.
- Compact Size: At 0.5 billion parameters, it offers a relatively small footprint for math-focused applications.
Performance Highlights
Validation on the fdyrd/MATH dataset shows an average accuracy of 16.6% across various difficulty levels and math subjects. On specific test sets, it achieved an accuracy of 28.6% on MATH500 and 38.2% on GSM8K.
Good For
- Applications requiring a dedicated, smaller model for mathematical reasoning.
- Research and development in specialized math-solving LLMs.
- Use cases where computational resources are limited but math capabilities are essential.