MiniMath-R1-1.5B
| Property | Value |
|---|---|
| Parameter Count | 1.5B |
| Model Type | Small Language Model |
| License | Apache 2.0 |
| Base Model | DeepSeek-R1-Distill-Qwen-1.5B |
| Training Data | oumi-ai/MetaMathQA-R1 |
What is MiniMath-R1-1.5B?
MiniMath-R1-1.5B is a specialized mathematical language model developed by Oumi AI, designed specifically for mathematical problem-solving. This model represents a significant achievement in compact AI models, reaching 44.4% accuracy on MMLU-Pro-Math - the highest performance among models with 1.5B parameters or fewer, marking a 6-point improvement over its base model.
Implementation Details
The model is implemented as a supervised fine-tune of DeepSeek-R1-Distill-Qwen-1.5B, utilizing the oumi-ai/MetaMathQA-R1 dataset. The training process was notably efficient, using 8 H100 GPUs for approximately 0.8 hours on Google Cloud Platform's us-east5 region, resulting in a minimal carbon footprint of just 0.07 kg.
- Optimized for mathematical reasoning and problem-solving
- Efficient training implementation with minimal environmental impact
- Built on proven DeepSeek architecture with specialized mathematical capabilities
Core Capabilities
- State-of-the-art mathematical problem solving for compact models
- Exposed thought process for educational applications
- Specialized focus on mathematical reasoning and computation
- Efficient deployment with minimal computational requirements
Frequently Asked Questions
Q: What makes this model unique?
MiniMath-R1-1.5B stands out for achieving the highest MMLU-Pro-Math accuracy (44.4%) among models of its size class, making it particularly effective for mathematical applications while maintaining a compact form factor.
Q: What are the recommended use cases?
The model is specifically designed for mathematical problem-solving and should be used as a conversational assistant for math-related tasks. It's particularly valuable in educational contexts where understanding the thought process is important. However, it should not be used for purposes outside of mathematics due to its specialized nature.
