Parameters
14B
Context Length
33K
Modality
Text
Architecture
Dense
KV Cache Rate
200 KB/tok
License
MIT
Release Date
30 Apr 2025
Knowledge Cutoff
Mar 2025
API Pricing (per 1M)
Self-hosted only
Precision
Context
Workload Memory: 10.58 GB
INT4 · 8K context
Base Weights: 6.52 GB
KV Cache: 1.56 GB
Runtime & Graphs: 2.50 GB
1x NVIDIA L4 24GB
10.6 / 21.6 GB usable
1x RTX 3060 12GB
10.6 / 10.8 GB usable
Apple M4 24GB
10.6 / 18.0 GB usable
Determine requirements for your specific workload:
Open in VRAM Calculator
Checkpoint format
FP16
CUDA
>= 11.8
PyTorch
>= 2.1
Transformers
>= 4.51.1
| Architecture | Precision | Attention kernel |
|---|---|---|
Blackwell, Hopper | Native FP16 | FlashAttention-3 |
Ada Lovelace, Ampere | Native FP16 | FlashAttention-2 |
Overall Rank
#115
Coding Rank
-
Phi-4 Reasoning Plus is an enhanced 14B parameter reasoning model by Microsoft trained with GRPO reinforcement learning for mathematical and scientific deduction. It generates explicit step-by-step reasoning traces across a 32K context window.
Attention
Attention Structure
Multi-Head Attention
Attention Heads
40
Key-Value Heads
10
Position Embedding
Absolute Position Embedding
RoPE Theta
500,000
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Base Precision
FP16
Hidden Dimension Size
5,120
Number of Layers
40
FFN Intermediate Size (Dense)
17,920
Tokenizer
Vocabulary Size
100,352
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.