Phi-2 Specs, Benchmarks & Downloads
Phi-2
Microsoft small language model
Status
active
Parameters
2.70
Context Window
2048
Modality
Text
Release Date
Dec 12, 2023
About
Explore Phi-2, a text AI model developed by Microsoft. View specifications, context window, architecture, license, supported platforms, features, benchmarks, official downloads, and related articles.
Phi-2 demonstrated strong reasoning capability despite its compact size.
Family
PhiDeveloper
MicrosoftArchitecture
TransformerLicense
Apache 2.0Hardware Requirements
Minimum:
Use Case: Local inference
GPU Minimum: GTX 1650
GPU VRAM: 4GB
System RAM: 8GB
Storage: 10GB
Operating System: Windows Linux
Phi-2 small model
Recommended:
Use Case: Desktop inference
GPU Minimum: RTX 3060
GPU VRAM: 12GB
System RAM: 16GB
Storage: 20GB
Operating System: Linux
Recommended local deployment
High Performance:
Use Case: Fast inference
GPU Minimum: RTX 4070
GPU VRAM: 12GB
System RAM: 32GB
Storage: 30GB
Operating System: Linux
High speed inference
Model Deployment Readiness
Model Ecosystem Support
18%
Platforms
33%
Features
1%
Downloads
20%
Platform Compatibility
33%
Supported across 4 of 12 platforms
Context Window
Industry-standard context window comparison
2,048
tokens
2K
4K
32K
128K
1M
2M
Database Rank
#90
Percentile
Top 11%
Capacity
Small
Benchmark Performance
Evaluation profile across publicly reported benchmarks
|
MMLU
|
Knowledge | 56.7 % | 2023 |
|
MATH-500
|
Mathematics | 41.4 % | 2023 |
|
GSM8K
|
Mathematics | 57.4 % | 2023 |
|
HellaSwag
|
Reasoning | 76.7 % | 2023 |
Highest Benchmark
HellaSwag
76.7%
Benchmarks Available
4
Latest Evaluation
Dec 2023
