ArchiveStudio/Phi-4-reasoning-plus

Overview

ArchiveStudio/Phi-4-reasoning-plus is a 14.7 billion parameter, dense decoder-only Transformer model developed by Microsoft Research. Fine-tuned from the Phi-4 base model, it leverages supervised fine-tuning on chain-of-thought traces and reinforcement learning to enhance its reasoning capabilities. This approach focuses on high-quality, advanced reasoning data, particularly in math, science, and coding skills, while also incorporating alignment data for safety and Responsible AI. The model has a context length of 32,768 tokens and is designed to provide detailed reasoning chains followed by concise solutions.

Key Capabilities

• Advanced Reasoning: Excels in complex reasoning tasks across math, science, and coding, outperforming many larger open-weight models.
• Chain-of-Thought (CoT) Generation: Produces detailed reasoning steps before presenting a solution, crucial for understanding complex problem-solving.
• Extended Context Handling: Demonstrates promising results with longer sequences, maintaining coherence and logical consistency over extended inputs, with experimental support up to 64K tokens.
• Optimized for Resource-Constrained Environments: Designed for use in memory/compute-constrained and latency-bound scenarios while delivering strong reasoning performance.
• Robust Safety Alignment: Incorporates a robust safety post-training approach via supervised fine-tuning and extensive red-teaming.

Good For

• Accelerating Research: Ideal as a building block for generative AI features and research on language models.
• Applications Requiring Logic & Reasoning: Suitable for general-purpose AI systems and applications (primarily in English) that demand strong logical processing.
• Educational Tools: Can be used to develop tools that require step-by-step problem-solving explanations in technical domains.
• Code Generation & Analysis: Shows strong performance in functional code generation and competitive coding benchmarks.