Phi-4-Mini
Parameters
3.8B
Context Length
128K
Modality
Text
Architecture
Dense
License
MIT
Release Date
27 Feb 2025
Knowledge Cutoff
Jun 2024
System Requirements
VRAM requirements for different quantization methods and context sizes
1,024 tokens
Consumer
1x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
128,000 tokens
Consumer
2x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
Architecture Diagram
Evaluation Benchmarks
Rank
#137
| Benchmark | Score | Rank |
|---|---|---|
General Knowledge | 0.673 | 32 |
Rankings
Overall Rank
#137
Coding Rank
-
About Phi-4-Mini
Microsoft Phi-4-Mini is a lightweight, open model from the Phi-4 family, engineered to operate efficiently in resource-constrained environments. This model is constructed from a combination of high-quality synthetic data and filtered public web content, with a particular emphasis on data dense in reasoning. Its core architecture is a dense, decoder-only Transformer, optimized with techniques such as grouped-query attention (GQA) and LongRoPE positional encoding to enhance inference speed and manage extended context lengths. The model incorporates an expanded vocabulary of 200,064 tokens, facilitating broad multilingual support.
Key advancements in Phi-4-Mini include an enhancement process that integrates supervised fine-tuning (SFT) and direct preference optimization (DPO), along with Reinforcement Learning from Human Feedback (RLHF) for robust instruction adherence and safety measures. This training methodology enables the model to exhibit strong reasoning capabilities, particularly in mathematical and logical tasks, and supports advanced functions such as function calling. The design prioritizes computational efficiency and low-latency performance, making it suitable for deployment in scenarios where memory and processing power are limited.
The intended use cases for Phi-4-Mini span general-purpose AI systems and applications that require strong reasoning in memory or compute-constrained environments, or those with latency-bound requirements. It is designed to accelerate research in language models and serve as a foundational building block for generative AI features. The model's compact size and optimized architecture allow for deployment on edge devices, including various mobile operating systems, by leveraging tools such as Microsoft Olive and the ONNX GenAI Runtime.
Technical Specifications
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
24
Key-Value Heads
8
Attention Head Dimension
-
Position Embedding
ROPE
RoPE Theta
10,000
Sliding Window Attention
Yes
Sliding Window Size
262,144
Sliding Window Ratio
-
Linear Attention
-
Linear Attention Ratio
-
Normalization
RMS Normalization
Activation Function
Swish
Dimensions
Hidden Dimension Size
3,072
Number of Layers
32
FFN Intermediate Size (Dense)
8,192
Multi-Token Prediction Heads
-
Tokenizer
Vocabulary Size
200,064
About Phi-4
The Microsoft Phi-4 model family comprises small language models prioritizing efficient, high-capability reasoning. Its development emphasizes robust data quality and sophisticated synthetic data integration. This approach enables enhanced performance and on-device deployment capabilities.