华为公开AI模型新专利,能够减少处理延时
Qi Cha Cha·2025-09-05 09:41

Core Viewpoint - Huawei has published a new patent for an AI model that aims to reduce processing latency by optimizing the handling of embedding vectors [1] Group 1: Patent Details - The patent titled "Method, Device, Program Product, and Storage Medium for Running AI Models" was published on September 5 [1] - The patent involves a host that includes a processor and is connected to a computing card, focusing on the management of input data sets [1] - The method allows the processor to identify first data from a second data group that is not present in a first data group, facilitating efficient data handling [1] Group 2: Technical Implications - The approach described in the patent enables the pre-fetching of embedding vectors to reduce latency caused by data transfer [1] - By processing the second data group on the computing card while utilizing information from the first embedding vector, the system enhances overall processing efficiency [1]