Loongson Technology: Loongson processor successfully runs DeepSeek large model

On February 7th, Jingshang Data reported that the device equipped with Loongson 3 CPU successfully started and operated the DeepSeek R1 7B model, achieving localized deployment with excellent performance and cost-effectiveness, and providing users with faster, stronger, and more economical training and inference experience. Recently, Loongson, together with industry partners such as Taichi Universe and others, completed the adaptation of the DeepSeek-R1 series models on the Taichi T100 acceleration card in just 2 hours, and quickly launched multiple large-scale models services including DeepSeek-R1-Distill-Qwen-7B. Currently, Loongson is actively working with Taichi Universe, Cambricon, Tiansu Zhixin, Alnolgy, openEuler and other partners to build a multi-form inference platform for the DeepSeek series models, helping enterprise users achieve intelligent transformation.