Phi-4
Parameters
14B
Context Length
16K
Modality
Text
Architecture
Dense
License
MIT License
Release Date
13 Dec 2024
Knowledge Cutoff
Nov 2024
System Requirements
VRAM requirements for different quantization methods and context sizes
1,024 tokens
Consumer
2x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
16,000 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
#135
| Benchmark | Score | Rank |
|---|---|---|
General Knowledge | 0.848 | 15 |
Professional Knowledge | 0.7 | 63 |
General Text | 1256 | 148 |
Rankings
Overall Rank
#135
Coding Rank
-
About Phi-4
Microsoft Phi-4 is a 14 billion parameter decoder-only Transformer model, developed as the latest iteration in Microsoft's series of small language models (SLMs). The model's primary objective is to deliver advanced reasoning capabilities efficiently, enabling deployment in environments with limited compute and memory, and for latency-sensitive applications. Phi-4 is designed to handle complex logical and mathematical tasks, along with general language processing, by focusing on the quality of its training data rather than solely on model scale.
A key innovation in Phi-4's architecture and training methodology lies in its strategic use of high-quality synthetic data, which constitutes a significant portion of its training corpus. This synthetic data, generated using techniques such as multi-agent prompting, instruction reversal, and self-revision workflows, is complemented by meticulously curated organic data from web content, academic books, and code repositories. This approach enables Phi-4 to acquire strong reasoning and problem-solving abilities, often surpassing models with larger parameter counts. The model's architecture retains a similar structure to its predecessor, Phi-3, but includes enhancements such as an extended context length.
Phi-4 supports a 16,000-token context length, allowing it to process and generate extensive long-form content. Its design prioritizes efficiency and robust performance in tasks requiring logical deduction, code generation, and scientific understanding. The model is intended for research and development, serving as a foundational component for generative AI features in various applications, particularly those demanding strong reasoning in resource-constrained or low-latency scenarios.
Technical Specifications
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
24
Key-Value Heads
8
Attention Head Dimension
-
Position Embedding
ROPE
RoPE Theta
250,000
Sliding Window Attention
No
Sliding Window Size
-
Sliding Window Ratio
-
Linear Attention
-
Linear Attention Ratio
-
Normalization
RMS Normalization
Activation Function
Swish
Dimensions
Hidden Dimension Size
3,072
Number of Layers
40
FFN Intermediate Size (Dense)
17,920
Multi-Token Prediction Heads
-
Tokenizer
Vocabulary Size
100,352
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.