DeepSeek Discloses Tech Details, Counters Cost Misrepresentation Claims
深度求索开源技术细节，驳斥成本造假指控
Chinese AI start-up DeepSeek disclosed technical details of its low-cost, high-performance models, refuting allegations of cost misrepresentation and gaining support from the open-source community. The company also fulfilled its promise to release five open-source AI infrastructure projects this week, launching FlashMLA and DeepEP to optimize chip performance for efficient model training and inference. Stephen Pimentel, CTO of San Francisco-based Dragonscale Industries, stated on X that DeepSeek’s open-source efforts effectively counter claims of dishonesty about its training procedures.
中国人工智能初创公司深度求索（DeepSeek）公开了其低成本、高性能模型的技术细节，驳斥了关于其成本造假的指控，并赢得了开源社区的支持。该公司还兑现了承诺，本周开始发布五个开源AI基础设施项目，目前已发布FlashMLA和DeepEP两个项目，旨在优化芯片性能以实现高效模型训练和推理任务。旧金山AI解决方案提供商Dragonscale Industries的首席技术官Stephen Pimentel在X上表示，深度求索通过开源技术有效反驳了对其训练过程的质疑。
