Openintelligent123/Phi-4-reasoning-plus
Openintelligent123/Phi-4-reasoning-plus is a 14.7 billion parameter dense decoder-only Transformer model developed by Microsoft Research. It is fine-tuned from Phi-4 using supervised fine-tuning and reinforcement learning, specifically optimized for advanced reasoning in math, science, and coding tasks. With a 32K token context length, it excels in scenarios requiring deep, multi-step reasoning and logical consistency, particularly in memory/compute-constrained and latency-bound environments.
Loading preview...
Model Overview
Openintelligent123/Phi-4-reasoning-plus is a 14.7 billion parameter language model developed by Microsoft Research, fine-tuned from the Phi-4 base model. It leverages supervised fine-tuning on a dataset of chain-of-thought traces and reinforcement learning, with a focus on high-quality data for advanced reasoning in math, science, and coding. The model is designed to generate responses with a distinct reasoning chain-of-thought block followed by a summarization block.
Key Capabilities & Features
- Advanced Reasoning: Specifically optimized for complex reasoning and logic tasks in mathematics, science, and coding.
- Chain-of-Thought (CoT) Generation: Produces detailed reasoning processes before providing solutions, enhancing transparency and accuracy.
- Extended Context: Supports a 32K token context length, with experimental results showing promising coherence up to 64K tokens for multi-step reasoning.
- Robust Safety Alignment: Incorporates supervised fine-tuning with Microsoft safety guidelines and undergoes extensive red-teaming for safety evaluation.
- Performance: Outperforms significantly larger open-weight models on various reasoning benchmarks like AIME, OmniMath, and GPQA-Diamond, and shows strong generalization to algorithmic problems like 3SAT and TSP.
Ideal Use Cases
- Research Acceleration: Serves as a building block for generative AI features and research in language models.
- Resource-Constrained Environments: Suitable for applications with memory/compute limitations.
- Latency-Bound Scenarios: Designed for use cases where response time is critical.
- Educational & Problem-Solving Tools: Excellent for applications requiring detailed, step-by-step reasoning for complex problems.