Openintelligent123/Phi-4-mini-reasoning
Openintelligent123/Phi-4-mini-reasoning is a 3.8 billion parameter, decoder-only Transformer model from the Phi-4 family, developed by Microsoft. Optimized for mathematical reasoning, it excels at multi-step, logic-intensive problem-solving tasks. This model is specifically fine-tuned with synthetic math data to deliver high-quality, step-by-step solutions in compute-constrained environments, supporting a 128K token context length.
Loading preview...
Model Overview
Phi-4-mini-reasoning is a 3.8 billion parameter model from the Microsoft Phi-4 family, specifically designed for advanced mathematical reasoning. It is a compact, decoder-only Transformer model built upon high-quality, reasoning-dense synthetic data, further fine-tuned for enhanced math problem-solving capabilities. The model supports an extensive 128K token context length, making it suitable for complex, multi-step logical tasks.
Key Capabilities
- Mathematical Reasoning: Excels at multi-step, logic-intensive mathematical problem-solving, including formal proof generation, symbolic computation, and advanced word problems.
- Efficiency: Optimized for memory/compute constrained environments and latency-bound scenarios, making it suitable for edge or mobile deployments.
- Context Maintenance: Designed to maintain context across multiple steps in complex reasoning tasks.
- Synthetic Data Training: Trained on over one million diverse synthetic math problems, generated and verified by a more capable model (Deepseek-R1), ensuring high-quality reasoning trajectories.
Performance Highlights
Phi-4-mini-reasoning demonstrates strong performance on reasoning benchmarks, achieving 57.5 on AIME, 94.6 on MATH-500, and 52.0 on GPQA Diamond. These scores indicate that this 3.8B-parameter model can achieve reasoning abilities comparable to much larger models, particularly in its specialized domain.
Intended Use Cases
This model is primarily intended for applications requiring deep analytical thinking and accurate, reliable solutions in mathematical domains. Potential use cases include educational applications, embedded tutoring systems, and lightweight deployment on edge devices where computational resources are limited. It is important to note that the model is designed and tested for math reasoning only and may exhibit factual incorrectness for general knowledge tasks due to its size, which can be mitigated by augmenting with a search engine in RAG settings.