[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-google-deepmind-decoupled-diloco-cross-datacenter-50pct":3,"news-related-99b8aabd-0fbe-428f-8f44-4cd2ab5b8066":31},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":24,"published_at":25,"created_at":26,"modified_at":27,"is_published":28,"publish_type":29,"image_url":13,"view_count":30},"99b8aabd-0fbe-428f-8f44-4cd2ab5b8066","Google DeepMind发布Decoupled DiLoCo：跨数据中心分布式训练的新突破","\nGoogle DeepMind于4月23日发布了Decoupled DiLoCo，这是一项结合Pathways异步编排系统与DiLoCo低通信训练方法的分布式训练架构创新，旨在解决超大规模AI模型跨地理分布训练的核心瓶颈。\n\nDecoupled DiLoCo的核心设计思路是\"解耦\"。Pathways系统负责协调异构芯片以独立速度运行，而DiLoCo专注于最小化跨数据中心通信开销。两者结合后，内层优化可在本地完成，外层更新仅进行低频同步，将跨站点通信量降低至原来的八分之一。初步基准测试显示，在分布式设置中可减少高达50%的训练时间。\n\n这一技术突破对当前大模型训练面临的现实挑战具有直接意义。随着模型规模突破万亿参数量级，跨数据中心的互连带宽和\"掉队者效应\"成为训练效率的主要瓶颈。Decoupled DiLoCo通过异步协调和极低带宽需求，使得地理分散的硬件资源能够高效协作训练同一个模型。\n\n从技术生态角度看，该架构支持GPU、TPU甚至边缘设备的混合部署，无需频繁数据交换。这为数据主权合规场景（如GDPR要求下的本地化处理）提供了可行的技术路径，同时也为算力资源不足的地区参与前沿模型训练降低了门槛。\n\n分布式训练效率的提升将直接影响大模型的训练成本和迭代速度，这一方向的持续创新对整个AI行业的基础设施建设至关重要。","https:\u002F\u002Fx.com\u002FGoogleDeepMind\u002Fstatus\u002F2047330987353239925","4d11edad-2df6-45f6-b71f-70f65de7f7fd",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"8cf7490f-2449-4ba7-be19-61befa0d92b4","google",{"id":18,"name":19,"slug":19,"description":13,"color":13},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":21,"name":22,"slug":22,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[],"google-deepmind-decoupled-diloco-cross-datacenter-50pct","2026-04-24T02:30:00Z","2026-04-23T22:07:46.541869Z","2026-08-19T02:08:40.142862Z",true,"agent",126,{"items":32},[33,38,43,48,53,58],{"id":34,"title":35,"news_slug":36,"published_at":37},"1e1d16e0-6505-4e09-9d59-11476d55bdea","Google全面拥抱Gemini搜索：一次对人类信息获取方式的根本性重构","google-gemini-search-replaces-web","2026-05-31T14:10:00+00:00",{"id":39,"title":40,"news_slug":41,"published_at":42},"183fb3be-e062-47e7-9591-7c2372e116c1","LLM 蒸馏的显存瓶颈不只在教师模型：离线 Top-K 与分块 KL 把长上下文训练装回单卡","llm-distillation-offline-top-k-chunked-kl","2026-08-05T20:08:13+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"217f417d-1b9c-475b-99f4-e21e7c909711","MHAR 把 Transformer 残差流从「单车道」拆成 H 条独立路由:子空间第一次有权自己挑历史层","multi-head-attention-residuals-mhar","2026-08-01T07:30:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"e77ca785-1f1e-4b09-b28f-6723c4115e56","Chrome 动态补丁要让浏览器不重启也能打补丁：LLM 把\"漏洞太多\"逼成了架构问题","chrome-dynamic-patching-llm-vulnerability","2026-08-01T06:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"f637e5a0-5e18-4ced-9aa4-2ce5df798a9c","Gemini 接管 Chrome 漏洞流水线:1072 个 bug、13 年陈年沙箱逃逸,LLM 重塑浏览器安全","gemini-chrome-vulnerability-pipeline","2026-07-31T10:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"a1ab01f3-ef5b-4240-aa99-7738f48591aa","PReM 用「按需刷新」撕开 LLM 长上下文压缩天花板:阿里团队 32K 上下文做到 16×\u002F32× 压缩仍保住多跳推理","prem-on-demand-refresh-32k","2026-07-18T20:08:00+00:00"]