input := input.trim();
Windows 10/11 or macOS 12+ (Monterey)
。关于这个话题，91吃瓜提供了深入分析
The simplest experiments have hundreds of hidden variables, none of which are in the eventual model—and for good reason—but our extraordinary ability to predict the behavior of the world depends on hidden, load-bearing walls and it's when those walls, those assumptions, begin to fail and break down that our models and our societies go with them.
A growing countertrend towards smaller (opens in new tab) models aims to boost efficiency, enabled by careful model design and data curation – a goal pioneered by the Phi family of models (opens in new tab) and furthered by Phi-4-reasoning-vision-15B. We specifically build on learnings from the Phi-4 and Phi-4-Reasoning language models and show how a multimodal model can be trained to cover a wide range of vision and language tasks without relying on extremely large training datasets, architectures, or excessive inference‑time token generation. Our model is intended to be lightweight enough to run on modest hardware while remaining capable of structured reasoning when it is beneficial. Our model was trained with far less compute than many recent open-weight VLMs of similar size. We used just 200 billion tokens of multimodal data leveraging Phi-4-reasoning (trained with 16 billion tokens) based on a core model Phi-4 (400 billion unique tokens), compared to more than 1 trillion tokens used for training multimodal models like Qwen 2.5 VL (opens in new tab) and 3 VL (opens in new tab), Kimi-VL (opens in new tab), and Gemma3 (opens in new tab). We can therefore present a compelling option compared to existing models pushing the pareto-frontier of the tradeoff between accuracy and compute costs.
（本报记者李林蔚、王明峰整理）