Jin10 data reported on July 10, this morning, Microsoft open sourced the latest version of the Phi-4 family, Phi-4-mini-flash-reasoning, on its official website. The mini-flash version continues the Phi-4 family’s characteristics of small parameters and strong performance, specifically designed for scenarios limited by Computing Power, memory, and latency, capable of running on a single GPU, suitable for edge devices like laptops and tablets. Compared to the previous version, mini-flash utilizes Microsoft’s self-developed innovative architecture, SambaY, resulting in a big pump in inference efficiency by 10 times, with average latency reduced by 2-3 times, achieving a significant improvement in overall inference performance.
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
Microsoft Open Source New Version of Phi-4: Inference Efficiency Rises 10 Times, Can Run on Laptops
Jin10 data reported on July 10, this morning, Microsoft open sourced the latest version of the Phi-4 family, Phi-4-mini-flash-reasoning, on its official website. The mini-flash version continues the Phi-4 family’s characteristics of small parameters and strong performance, specifically designed for scenarios limited by Computing Power, memory, and latency, capable of running on a single GPU, suitable for edge devices like laptops and tablets. Compared to the previous version, mini-flash utilizes Microsoft’s self-developed innovative architecture, SambaY, resulting in a big pump in inference efficiency by 10 times, with average latency reduced by 2-3 times, achieving a significant improvement in overall inference performance.