[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-α-entmax":3},{"tag":4,"articles":9},{"id":5,"name":6,"slug":6,"article_count":7,"description_zh":8,"description_en":8},"5c7f114c-bcd2-4cdc-8411-b1e36f15f4d5","α-entmax",0,null,[10],{"id":11,"slug":12,"title":13,"summary":14,"category":15,"image_url":16,"cover_image":16,"language":17,"created_at":18},"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"]