Phi
Microsoft's small language model series, designed to be efficient enough to run on phones and edge devices.
Think of it like
Proof that size isn't everything; clever design can make tiny models surprisingly capable.
Example
Phi-2 (2.7B) performs comparably to much larger models on reasoning tasks; enables on-device AI apps.
How it actually works
Phi uses a technique called "textbook quality" data selection: curated, high-quality examples rather than massive raw web data. Result is strong performance-per-parameter.
For product teams
On-device inference means better privacy, latency, and cost; huge for mobile and edge AI.
For engineers
Phi trades off breadth for depth; very good at reasoning, weaker on factual recall.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome