[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-sparse-attention":3},{"tag":4,"articles":10},{"id":5,"name":6,"slug":7,"article_count":8,"description_zh":9,"description_en":9},"550a3d47-e632-4452-b9c2-c89caea4766e","sparse attention","sparse-attention",0,null,[11],{"id":12,"slug":13,"title":14,"summary":15,"category":16,"image_url":17,"cover_image":17,"language":18,"created_at":19},"475844e6-3e2c-49a6-aea0-86a94945d2c2","dashattention-differentiable-adaptive-sparse-attention-zh","DashAttention 讓稀疏長上下文可微","DashAttention 把長上下文的分層稀疏注意力改成可微、可自適應的選擇機制，讓模型在 75% 稀疏下仍能維持接近全注意力的表現。","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779171840613-dq1r.png","zh","2026-05-19T06:23:32.886786+00:00"]