Phi-3
#9 in Lütte & Edge-Modellemicrosoft · v3 · siet 2024-04-23 · 10× · tolest 30. Juni 2026
25
Momentum
Phi-3 is a family of small open-source language models (SLMs) from Microsoft, comprising Phi-3-mini (3.8B), Phi-3-small (7B), Phi-3-medium (14B), and the multimodal Phi-3-vision (4.2B). The models are trained on carefully curated high-quality synthetic data to maximize reasoning performance with minimal parameter count. They are designed for resource-constrained and edge/on-device environments and can be run locally via Ollama, Azure AI Studio, or Hugging Face. Models support up to 128K token context and are released under the MIT license.
Momentum-Verloop
04.04.03.07.
Features
| Key Benchmark (%) | MMLU (5-Shot): Phi-3-mini 68.8%; Phi-3-small 75.7%; Phi-3-medium 78.0% |
| Context Window (Tokens) | 4K or 128K tokens (mini); Small: 8K/128K; Medium: 4K/128K |
| License | MIT License (open weights) |
| Multimodality | Text (mini/small/medium); Phi-3-vision (4.2B) additionally image+text |
| Platform | Azure AI Foundry/Studio, Hugging Face, Ollama, ONNX Runtime, NVIDIA NIM |
| Release Date | April 23, 2024 (Phi-3-mini); Phi-3-small/medium/vision: May 21, 2024 |