[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-fine-tuning":3},{"tag":4,"articles":10},{"id":5,"name":6,"slug":6,"article_count":7,"description_zh":8,"description_en":9},"3725f3eb-5764-4f52-a203-bdfaacae8acc","fine-tuning",4,"微調是把通用模型改造成特定任務工具的關鍵步驟，常見於新詞注入、指令對齊與多模態適配。重點不只在訓練技巧，也在初始化、資料分佈、VRAM 需求與語言覆蓋，直接影響生成品質與部署成本。","Fine-tuning adapts a base model to a narrower task or domain, from seeding new vocabulary and aligning instruction behavior to adapting vision-language models. The practical issues are initialization, data quality, VRAM limits, and language coverage, all of which shape output quality and deployment cost.",[11,20,27,35],{"id":12,"slug":13,"title":14,"summary":15,"category":16,"image_url":17,"cover_image":17,"language":18,"created_at":19},"18fb2e62-3d41-4b4c-8d65-e91e5f20ea28","microsoft-goalcover-fine-tuning-gaps-en","Microsoft’s GoalCover finds fine-tuning gaps","Microsoft Research’s GoalCover spots missing capabilities in fine-tuning data before training, and improved Qwen-3-14B reward scores.","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778462450292-viev.png","en","2026-05-11T01:20:34.483926+00:00",{"id":21,"slug":22,"title":23,"summary":24,"category":16,"image_url":25,"cover_image":25,"language":18,"created_at":26},"05451495-1e4d-4e70-855f-f92e68a1a699","how-to-build-vintage-llm-testbed-5-steps-en","How to Build a Vintage LLM Testbed in 5 Steps","Build a 1930-cutoff LLM testbed to study historical reasoning and contamination-free generalization.","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777945253760-2l44.png","2026-05-05T01:40:33.098256+00:00",{"id":28,"slug":29,"title":30,"summary":31,"category":32,"image_url":33,"cover_image":33,"language":18,"created_at":34},"e031b580-6869-4e89-886d-f190e0adfa86","unsloth-qwen35-partial-fine-tuning-en","Unsloth Adds Part-by-Part Qwen3.5 Fine-Tuning","Unsloth now lets you fine-tune Qwen3.5 vision models by layer type, with faster training, lower VRAM, and 201-language support.","tools","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775218020650-71sw.png","2026-04-03T12:06:39.044523+00:00",{"id":36,"slug":37,"title":38,"summary":39,"category":40,"image_url":41,"cover_image":41,"language":18,"created_at":42},"e487e7c6-aa22-484d-9555-46261cc7a91d","grounded-token-initialization-new-vocabulary-en","A Better Way to Seed New LM Tokens","GTI grounds new vocabulary tokens before fine-tuning, aiming to preserve distinctions that mean initialization tends to collapse.","blockchain","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775196588405-1a7u.png","2026-04-03T06:09:29.832749+00:00"]