Iambackup/Phi-4-reasoning-plus
Iambackup/Phi-4-reasoning-plus is a 14.7 billion parameter dense decoder-only Transformer model developed by Microsoft Research. Fine-tuned from Phi-4, it specializes in advanced reasoning, mathematical, scientific, and coding tasks through supervised fine-tuning on chain-of-thought traces and reinforcement learning. With a 32k token context length, it is optimized for memory/compute-constrained environments and latency-bound scenarios requiring strong logical processing.
Loading preview...
Model Overview
Iambackup/Phi-4-reasoning-plus is a 14.7 billion parameter model from Microsoft Research, fine-tuned from the Phi-4 architecture. It leverages supervised fine-tuning on chain-of-thought traces and reinforcement learning to excel in complex reasoning tasks, particularly in math, science, and coding. The model is designed to provide detailed reasoning chains followed by concise solutions.
Key Capabilities
- Advanced Reasoning: Demonstrates strong performance on challenging reasoning benchmarks like AIME, OmniMath, and GPQA-Diamond.
- Code Generation: Achieves high scores on LiveCodeBench and HumanEvalPlus, indicating proficiency in functional code generation.
- Instruction Following: Shows significant improvement in instruction following (IFEval, ArenaHard) compared to its base model.
- Context Handling: Supports a 32k token context length, with experimental results showing coherence up to 64k tokens for deep, multi-step reasoning.
- Optimized for Efficiency: Suitable for memory/compute-constrained and latency-bound environments.
Intended Use Cases
- Research Acceleration: Ideal for advancing research in language models and as a building block for generative AI features.
- Reasoning and Logic Applications: Excels in scenarios requiring systematic thinking, analysis, and problem-solving.
- Educational Tools: Can be used for generating explanations and solutions for math, science, and coding problems.
Usage Notes
For optimal performance, inference should use specific parameters: temperature=0.8, top_k=50, top_p=0.95, and do_sample=True. A specific ChatML system prompt is required to engage its chain-of-thought reasoning capabilities.