boh69/Phi-4-mini-instruct

TEXT GENERATIONConcurrent Unit Cost:1Model Size:3.8BQuant:BF16Context Size:32kPublished:Jul 7, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

Phi-4-mini-instruct is a 3.8 billion parameter instruction-tuned decoder-only Transformer model developed by Microsoft, part of the Phi-4 family. It is built upon synthetic data and filtered public websites, with a strong focus on high-quality, reasoning-dense data. The model supports a 128K token context length and excels in memory/compute-constrained environments, latency-bound scenarios, and strong reasoning tasks, particularly in math and logic.

Loading preview...

Model Overview

Phi-4-mini-instruct is a 3.8 billion parameter instruction-tuned language model from the Phi-4 family, developed by Microsoft. It features a dense decoder-only Transformer architecture with a 128K token context length and a 200K vocabulary, incorporating grouped-query attention and shared input/output embedding for efficiency. The model was trained on 5 trillion tokens of synthetic and filtered public data, with a strong emphasis on high-quality, reasoning-dense content, including math, coding, and common sense reasoning.

Key Capabilities

  • Strong Reasoning: Demonstrates robust performance in reasoning tasks, especially in mathematics and logic, as evidenced by high scores on benchmarks like GSM8K (88.6) and MATH (64.0).
  • Instruction Adherence & Safety: Enhanced through supervised fine-tuning and direct preference optimization to ensure precise instruction following and robust safety measures.
  • Multilingual Support: Features a larger vocabulary and improved post-training techniques for multilingual capabilities, supporting languages such as Arabic, Chinese, English, French, German, Japanese, and Spanish.
  • Tool-enabled Function Calling: Supports a specific input format for function calling, allowing users to provide tools in JSON format within the system prompt.

Ideal Use Cases

  • Resource-Constrained Environments: Optimized for deployment in memory or compute-constrained settings.
  • Low-Latency Applications: Suitable for scenarios where quick response times are critical.
  • General Purpose AI Systems: Can serve as a foundational building block for various generative AI features and applications requiring strong reasoning abilities.
  • Research Acceleration: Designed to accelerate research in language and multimodal models.

Finetunes
1 models