[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-centroid-masking":3},{"tag":4,"articles":10},{"id":5,"name":6,"slug":7,"article_count":8,"description_zh":9,"description_en":9},"c2c0e2c2-3af0-4a91-b170-38dfe6ffb7f0","centroid masking","centroid-masking",0,null,[11],{"id":12,"slug":13,"title":14,"summary":15,"category":16,"image_url":17,"cover_image":17,"language":18,"created_at":19},"fe630502-5455-4001-a6bf-0643f9eb469d","gemma-4-assistant-models-faster-draft-tokens-zh","Gemma 4 助手模型加速草稿 Token","Gemma 4 的 E2B 與 E4B 助手模型用 centroid masking，把草稿 token 的 lm_head 計算量砍到約 45 倍，且品質損失很小。","tools","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778278246167-hskc.png","zh","2026-05-08T22:10:33.309766+00:00"]