Model Family
Phi
Microsoft's small-but-mighty model family trained on heavily-curated synthetic data.
Definition
Microsoft's Phi series (Phi-1 → Phi-4) shows that careful data curation — filtered web text plus synthetic textbook-quality examples — lets small models (1–14B parameters) match larger ones on many benchmarks. Phi targets the on-device and edge inference niche.
Common use cases
- On-device
- Edge
- Embedded agents