开源推荐 None GitHub 热门项目：gemma-2B-10M GitHub项目：gemma-2B-10M 仓库地址：https://github。com/mustafaaljadery/gemma-2B-10M Stars：933 | 作者：mustafaaljadery 项目描述：Gemma 2B the https attention 2026-06-29 GitHub Trending
智能AI None 高斯混合注意力：通过概率潜在路由进行线性时间序列混合 18283v1 Announce Type: new Abstract: The dense token-to-token interaction pattern of standard dot-product attention rema and attention GMA 2026-06-18 Yongchao Huang, Hassan Raza
开发者生态 None 变压器注意力执行控制不足 【HN用户评论摘要】 > they lack an explicit architecture for the executive control of attention found in humansDeceptive terminol the with that 2026-06-11 derbOac
智能AI None 模糊窗口注意 arXiv:2606.09862v1 Announce Type: new Abstract: The Softmax Attention operation in Transformer language models has a qua the and Attention 2026-06-10 Axel Laborieux, Christos Sourmpis, Juan Gabriel Kostelec, Qinghai Guo
智能AI None 超越TurboQuant！OSCAR：面向真实服务的2-bit KV Cache量化 作者 | Zhongzhu Zhou, Donglin Zhuang, Jisen Li, Ziyan Chen, Shuaiwen Leon Song, Ben Athiwaratkun, Xiaoxia Wu 从 KV Cache 瓶颈 OSCAR attention bit 2026-05-29 Zhongzhu Zhou等
智能AI None 老黄喝豆汁「破防」背后，国产GPU正在填上CUDA护城河 新智元报道 【新智元导读】 老黄在北京喝豆汁「翻车」，全网笑疯了。但真正值得警惕的，是他背后那个正在长出来的「中国版CUDA生态」。从万卡集群到机器狗，从SGLang主线到AI Agent自动迁移，这家公司这次不只是秀芯片，而是在重写国产G MUSA CUDA GPU 2026-05-20 新智元
智能AI None 视觉语言模型中的可靠性所在：注意力、隐藏状态和因果回路的机制研究 arXiv:2605.08200v1 Announce Type: new Abstract: A pervasive intuition holds that vision-language models (VLMs) are most attention and the 2026-05-12 Logan Mann, Ajit Saravanan, Ishan Dave, Shikhar Shiromani, Saadullah Ismail, Yi Xia, Emily Huang