钟无艳

Covid still int'l health emergency: WHO_我的网站

一 |     Geneva, Oct 20 (UNI) The World Health Organization (WHO) said that Covid-19 is still a Public Health Emergency of International Concern (PHEIC), the WHO's highest alert level.



The announcement came although the number of weekly deaths is almost at its lowest level since the pandemic began.



The WHO's International Health Regulations Emergency Committee said on Wednesday after its quarterly assessment meeting last week that despite a decline in severe cases of Covid-19, and the falling number of weekly deaths, deaths from Covid-19 nevertheless remain high compared with other respiratory viruses.



It also warned of Covid-19-related complications and post-Covid-19 conditions, with the full impact of these still not completely understood. The outbreak could also evolve during the upcoming winter season in the Northern Hemisphere, the committee said.



Meanwhile, the current gaps in global surveillance of Covid-19 have hindered early identification and evaluation of the virus' evolution. With the virus expected to continue evolving, the committee said that the genetic and antigenic characteristics of future variants cannot yet be reliably predicted. Evolving variants may pose challenges to current vaccines and therapeutics, the committee warned.



"Given the above considerations, the committee concurred that continued coordination of the international response is necessary," and considered that "the situation remains dynamic and requires frequent reassessments, and that the termination of the PHEIC, when considered feasible, should be implemented as safely as possible."



The committee recommended that there should be three key priorities in the future: strengthening surveillance and achieving vaccination targets for at risk-groups; continuing to increase access to affordable therapeutics; and strengthening pandemic preparedness planning, while continuing to protect the most at-risk groups.

UNI/XINHUA PRT。

二 |     

凤凰网科技讯 8月24日，摩尔线程正式发布《MTT S5000 Prefill-as-a-Service技术白皮书》，基于旗舰级AI训推一体智算卡MTT S5000，提出Prefill-Decode异构解耦方案，面向AI Agent、代码生成、超长文档分析等长上下文推理场景，旨在破解推理成本瓶颈。白皮书指出，大模型输入上下文规模进入百K至1M级别后，Prefill（预填充，计算密集型）与Decode（解码，访存密集型）在同一资源池混跑造成结构性算力错配。解决方案是将二者彻底解耦，分别运行在各自最契合的硬件资源池上，实现单位Token基础设施成本大幅下降。

三 | MTT S5000硬件特性与Prefill任务高度契合：提供高稠密算力压缩首字时延，原生支持FP8全精度计算，并全面兼容CUDA及PyTorch、vLLM等主流推理框架。实测数据显示，在64K上下文场景下，单机Prefill吞吐达95,920 tok/s，按智谱API定价测算，满负载下单机月收入可达64.6万元；400K极限长序列场景TPM达2.6MTokens，吞吐平稳可控。摩尔线程表示，大模型产业竞争下半场本质上是推理效率与商业回报的较量，该方案旨在帮助企业在万亿参数与Agent时代降低总拥有成本。完整白皮书已在官网发布。

四 |

Current article:http://3uknz.linxiuhuaishuangniuzhenrua.bond/xa8qy2p/20260826/7590.html

Published on:20:30:57