[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-cold-start":3},{"tag":4,"articles":9},{"id":5,"name":6,"slug":6,"article_count":7,"description_zh":8,"description_en":8},"b28b8990-e6c1-4e93-a0e3-50cfb33fd2ef","cold-start",1,null,[10],{"id":11,"slug":12,"title":13,"summary":14,"category":15,"image_url":16,"cover_image":16,"language":17,"created_at":18},"8e6e5e5b-c51f-495e-a596-203fb64c71eb","tsallis-loss-reasoning-model-training-zh","Tsallis loss 讓推理模型更快脫困","這篇論文用 Tsallis q-logarithm 搭出一條損失函數光譜，想解決推理模型在冷啟動時卡住的問題。它把 RLVR 和 latent trajectory 的 log-marginal-likelihood 串成可調參的連續體。","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777443006073-083j.png","zh","2026-04-29T06:09:37.277494+00:00"]