DeepSeek Unveils NSA Tech to Boost AI Model Efficiency
DeepSeek推出NSA技术,优化AI模型计算效率
DeepSeek has outlined its next development priorities in a new technical study co-authored by CEO Liang Wenfeng and 14 other researchers, focusing on “Native Sparse Attention” (NSA)—a technique aimed at enhancing AI model efficiency in processing vast amounts of data.
DeepSeek在一项新技术研究中透露了其下一步发展重点,该研究由创始人兼CEO梁文峰等15名作者共同撰写,探讨了一种名为“原生稀疏注意力”(NSA)的技术,旨在提升AI模型在处理海量数据时的效率。
The study highlights how Liang and his team continue to push the boundaries of AI research, following their breakthrough open-source AI models, V3 and R1, which were developed at a fraction of the cost and computing power compared to major tech firms’ LLM projects.
研究显示,DeepSeek团队在其开源AI模型V3和R1取得突破后,仍在推动更高效、低成本的大语言模型(LLM)研发。
