就在昨天，Kimi 和 Deepseek 分别发布了一篇论文，发布时间和论文主题内容都高度相似。我第一时间把两篇论文阅读完了，急忙跑来写写我对这两篇论文的感想和体会。

Codesign真是妙不可言啊 Dynamic sparse attention的主要问题就是如果用flashattention style的kernel写法，那么对于每个query block，需要load每个token激活的key/value block的并集，在最坏条件下是会degrade到full attention的。

Modern life makes us tired, right? But research from societies in Africa and South America suggests people in the ancient ...

Modern life makes us tired, right? But research from societies in Africa and South America suggests people in the ancient ...

BEIJING, Feb. 20 (Xinhua) -- Chinese consumers have benefited from the country's pro-consumption policies, which cover goods such as mobile phones, home appliances and automobiles, said the Ministry ...

The best figure skaters in the world travel to Seoul, South Korea, where they will contest the 2025 edition of the Four Continents Championships.

租金回报率超过银行定存，吸引更多投资者买楼收租，过去两日最少录得5宗投资者入市个案，当中沙田第一城占3宗，以市值租金计算，租金回报率超过4厘。业界指出，政府积极引入外来人才，以及来港入读大学的内地生有增无减，租金走势向上，吸引更多投资者「买砖头」。

Stefano Ghisolfi重复攀爬了The Full Journey路线，这条难度为9b的线路位于西班牙Margalef地区，由Alexander Megso于2022年开辟。 Stefano Ghisolfi进行了The Full ...

JERUSALEM, Feb. 16 (Xinhua) -- Israeli Prime Minister Benjamin Netanyahu, speaking alongside U.S. Secretary of State Marco ...

BEIJING, Feb. 15 (Xinhua) — An article by Xi Jinping, general secretary of the Communist Party of China (CPC) Central Committee, on improving the systems for exercising full and rigorous Party ...

解放周末！用AI写周报又被老板夸了！点击这里，一键生成周报总结，无脑直接抄 → https://ai.sohu.com/pc/textHome?_trans_=030001_yljdaikj ...

Jeffrey Sachs: The protectionism will fail and will increasingly isolate the US in the world economy and politics. There are ...