Phi Series
A family of small language models developed by Microsoft Research, known for their cost-effectiveness. The models achieve performance approaching that of larger models at reduced parameter counts through techniques such as high-quality training data curation and knowledge distillation.
Timeline (last 90 days)
The Phi series is a family of small-parameter language models released by Microsoft Research since 2023, including versions such as Phi-1, Phi-1.5, Phi-2 and Phi-3, with parameter sizes ranging from 1.3 billion to 14 billion.
The core idea of the Phi series is to use textbook-level high-quality data so that models with smaller parameter counts approach the performance of large models on reasoning, math and code tasks.
Microsoft plays a complex role in the open-source model ecosystem: it is both an important supporter of the Llama series and demonstrates its own research capabilities through the Phi series.
All Facts (3)
The Phi series is a family of small-parameter language models released by Microsoft Research since 2023, including versions such as Phi-1, Phi-1.5, Phi-2 and Phi-3, with parameter sizes ranging from 1.3 billion to 14 billion.
50%UnverifiedThe core idea of the Phi series is to use textbook-level high-quality data so that models with smaller parameter counts approach the performance of large models on reasoning, math and code tasks.
50%UnverifiedMicrosoft plays a complex role in the open-source model ecosystem: it is both an important supporter of the Llama series and demonstrates its own research capabilities through the Phi series.
50%