[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-how-to-build-ai-research-foundations-with-deepmind-zh":3,"article-related-how-to-build-ai-research-foundations-with-deepmind-zh":30,"series-research-a9b25f93-fd42-4aa7-95be-c4e648ad48c7":82},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":22,"views":26,"created_at":27,"published_at":28,"topic_cluster_id":29},"a9b25f93-fd42-4aa7-95be-c4e648ad48c7","how-to-build-ai-research-foundations-with-deepmind-zh","怎麼用 DeepMind 建立 AI 研究基礎","\u003Cp data-speakable=\"summary\">這篇教你用 DeepMind 課程與本機實作，建立現代語言模型與微調的入門基礎。\u003C\u002Fp>\u003Cp>這篇給想理解現代語言模型、準備進入研究閱讀，或想把理論接到實作的開發者與資料\u003Ca href=\"\u002Fnews\u002Fgoogle-io-shift-ai-science-agents-zh\">科學\u003C\u002Fa>學習者看。照做完，你會得到一套可重複的學習流程，外加一個能跑起來的本機模型練習專案。\u003C\u002Fp>\u003Cp>你會一路完成課程導讀、Python \u003Ca href=\"\u002Fnews\u002Fopenai-anthropic-ai-jobs-doom-zh\">工作\u003C\u002Fa>區、核心概念整理、模型推論測試，最後做一次小型微調練習。課程素材來自 \u003Ca href=\"https:\u002F\u002Fwww.datacamp.com\u002Ftracks\u002Fgoogle-deepmind-ai-research-foundations\">Google DeepMind: AI Research Foundations\u003C\u002Fa>，研究脈絡則可參考 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\">Google DeepMind GitHub\u003C\u002Fa>。\u003C\u002Fp>\u003Ch2>開始之前\u003C\u002Fh2>\u003Cul>\u003Cli>DataCamp 帳號，且可存取 Google DeepMind: AI Research Foundations 追蹤課程\u003C\u002Fli>\u003Cli>Google 帳號，若你的工作區需要登入\u003C\u002Fli>\u003Cli>Python 3.10+\u003C\u002Fli>\u003Cli>Node 20+，僅在你要做 companion web demo 時需要\u003C\u002Fli>\u003Cli>JupyterLab 4+ 或 VS Code 1.85+\u003C\u002Fli>\u003Cli>Git 2.40+\u003C\u002Fli>\u003Cli>至少 8 GB RAM，建議 16 GB\u003C\u002Fli>\u003Cli>可選：NVIDIA GPU 與 CUDA 12+，用於本機實驗\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>Step 1: 開啟課程追蹤頁\u003C\u002Fh2>\u003Cp>目的：先把學習路線看完整，再開始寫任何程式。這樣你會知道哪些章節先讀、哪些章節後做，避免概念與實作順序打架。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779963521409-34ys.png\" alt=\"怎麼用 DeepMind 建立 AI 研究基礎\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>請登入 DataCamp，打開課程追蹤頁，快速掃過模組名稱與單元順序，並把會碰到語言模型、訓練、評估的章節記到你的筆記中。接著建立一份待辦清單，準備逐章完成。\u003C\u002Fp>\u003Cp>驗收：你應該看到課程首頁、章節列表，以及你自己的學習清單。\u003C\u002Fp>\u003Ch2>Step 2: 建立 Python 工作區\u003C\u002Fh2>\u003Cp>目的：準備一個乾淨環境，讓你在做 notebook 與套件安裝時可以重現結果。獨立環境能降低版本衝突，也方便你之後替換模型或資料集。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779963527914-bnud.png\" alt=\"怎麼用 DeepMind 建立 AI 研究基礎\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cpre>\u003Ccode>python3 -m venv .venv\nsource .venv\u002Fbin\u002Factivate\npython -m pip install --upgrade pip\npip install jupyterlab transformers datasets accelerate evaluate sentencepiece\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>如果你習慣 conda，也可以建立等效環境，只要 Python 版本維持在 3.10 以上即可。先保持套件精簡，等課程真的需要時再加裝其他依賴。\u003C\u002Fp>\u003Cp>驗收：你應該可以啟動 JupyterLab，並且順利匯入核心套件。\u003C\u002Fp>\u003Ch2>Step 3: 整理核心模型概念\u003C\u002Fh2>\u003Cp>目的：先建立腦中的模型，再進入微調程式。你需要先弄懂 tokens、embeddings、attention、pretraining、instruction tuning 與 evaluation，後面的課程才會串得起來。\u003C\u002Fp>\u003Cp>請在閱讀每個單元時，用自己的話寫下一句定義，再補上一個實務問題，例如 tokenization 如何影響 context length，或為\u003Ca href=\"\u002Fnews\u002Fwhy-verkor-turboquant-silicon-ip-matters-zh\">什麼\u003C\u002Fa> fine-tuning 會改變模型行為。這份筆記會變成你之後排查問題的索引。\u003C\u002Fp>\u003Cp>驗收：你應該能用白話講出語言模型從文字輸入到文字輸出的流程。\u003C\u002Fp>\u003Ch2>Step 4: 執行小型模型推論\u003C\u002Fh2>\u003Cp>目的：確認你的環境真的能載入預訓練模型並產生文字。這一步是後面訓練與微調之前的基準測試，能先排除安裝與硬體問題。\u003C\u002Fp>\u003Cp>先選一個小模型，例如 compact causal language model，然後只做三件事：載入 tokenizer、載入 model、產生短輸出。Notebook 不需要複雜，重點是把流程跑通。\u003C\u002Fp>\u003Cpre>\u003Ccode>from transformers import AutoTokenizer, AutoModelForCausalLM\n\nmodel_name = \"distilgpt2\"\ntokenizer = AutoTokenizer.from_pretrained(model_name)\nmodel = AutoModelForCausalLM.from_pretrained(model_name)\n\nprompt = \"Explain fine-tuning in one paragraph:\"\ninputs = tokenizer(prompt, return_tensors=\"pt\")\noutput = model.generate(**inputs, max_new_tokens=40)\nprint(tokenizer.decode(output[0], skip_special_tokens=True))\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>驗收：你應該在 notebook 看到一段生成文字，代表本機語言模型流程已經可用。\u003C\u002Fp>\u003Ch2>Step 5: 微調 toy dataset\u003C\u002Fh2>\u003Cp>目的：用一個小資料集練習完整研究流程，讓你看見資料準備、訓練參數與評估如何接在一起。這一步的重點不是追求高分，而是完成一次可追蹤的訓練迴圈。\u003C\u002Fp>\u003Cp>請選一個很小的文字資料集，跑短時間訓練，設定少量 epochs 或 steps，並把 checkpoint 存下來。接著記錄 loss 變化，再比較微調前後的輸出，觀察模型是否真的學到你的樣本特徵。\u003C\u002Fp>\u003Cp>驗收：你應該拿到一個已儲存的 checkpoint、一份訓練紀錄，以及微調前後可辨識的輸出差異。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>指標\u003C\u002Fth>\u003Cth>基準／優化前\u003C\u002Fth>\u003Cth>結果／優化後\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>模型行為\u003C\u002Ftd>\u003Ctd>通用型預訓練補全\u003C\u002Ftd>\u003Ctd>微調後的任務導向補全\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>訓練可視性\u003C\u002Ftd>\u003Ctd>沒有本機紀錄\u003C\u002Ftd>\u003Ctd>有 loss 曲線與 checkpoint\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>流程把握度\u003C\u002Ftd>\u003Ctd>只有概念理解\u003C\u002Ftd>\u003Ctd>完成端到端訓練練習\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>常見錯誤\u003C\u002Fh2>\u003Cul>\u003Cli>一開始就選太大的模型。修法：先用 distilgpt2 這類小模型驗證流程，再逐步升級。\u003C\u002Fli>\u003Cli>沒有隔離環境。修法：把所有練習放在 virtual environment，避免套件版本互相干擾。\u003C\u002Fli>\u003Cli>跳過概念章就直接訓練。修法：先完成 tokens 與 attention 的學習，再進入微調。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>接下來可以看什麼\u003C\u002Fh2>\u003Cp>完成基礎後，下一步可以做一個小型聊天介面、比較不同提示詞策略，或替特定領域資料做微調，讓研究概念變成可展示的作品。\u003C\u002Fp>","這篇教你用 DeepMind 的課程與本機實作，建立現代語言模型與微調的入門基礎。","www.datacamp.com","https:\u002F\u002Fwww.datacamp.com\u002Ftracks\u002Fgoogle-deepmind-ai-research-foundations",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779963521409-34ys.png","research","zh","a3c57be7-a302-4666-a308-113cb75f7494",[17,18,19,20,21],"DeepMind","DataCamp","Python","Transformers","fine-tuning",[23,24,25],"先用課程地圖建立學習順序，再開始實作。","用獨立 Python 環境和小模型，降低排錯成本。","完成一次 toy dataset 微調，才能把概念變成可重複流程。",3,"2026-05-28T10:17:24.309519+00:00","2026-05-28T10:17:24.296+00:00","0c35a120-52fc-41fc-afa3-d404eb934158",{"tags":31,"relatedLang":41,"relatedPosts":45},[32,34,36,37,39],{"name":19,"slug":33},"python",{"name":18,"slug":35},"datacamp",{"name":21,"slug":21},{"name":38,"slug":38},"transformers",{"name":17,"slug":40},"deepmind",{"id":15,"slug":42,"title":43,"language":44},"how-to-build-ai-research-foundations-with-deepmind-en","How to Build AI Research Foundations with DeepMind","en",[46,52,58,64,70,76],{"id":47,"slug":48,"title":49,"cover_image":50,"image_url":50,"created_at":51,"category":13},"f374155a-c29e-478c-b7a5-679cad1c51e4","crdts-keep-replicas-in-sync-without-locks-zh","CRDT 讓副本不用鎖也能同步","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781011086259-4p4k.png","2026-06-09T13:17:34.493426+00:00",{"id":53,"slug":54,"title":55,"cover_image":56,"image_url":56,"created_at":57,"category":13},"4b3b5a50-45b7-4238-a38b-160f82e323ff","post-deterministic-systems-autonomous-infra-zh","後決定性分散系：自治基礎設施新框架","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781010194792-5ogb.png","2026-06-09T13:02:32.717551+00:00",{"id":59,"slug":60,"title":61,"cover_image":62,"image_url":62,"created_at":63,"category":13},"04e45398-9814-4907-b416-fcb5b8d69508","causal-learnability-formal-language-tasks-zh","用因果法量化任務可學性","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780987696075-l4g0.png","2026-06-09T06:47:34.438642+00:00",{"id":65,"slug":66,"title":67,"cover_image":68,"image_url":68,"created_at":69,"category":13},"75bcc569-5e89-45c8-b809-6f169e929f4b","rl-training-hands-off-control-gradually-zh","RL 先接管再放手","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780986786312-03yo.png","2026-06-09T06:32:32.849589+00:00",{"id":71,"slug":72,"title":73,"cover_image":74,"image_url":74,"created_at":75,"category":13},"e3ecab4b-7cc7-4246-baf6-e1c170d86ca5","omnigamearena-vlm-game-agent-benchmark-zh","OmniGameArena 讓 VLM 遊戲代理更好比","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780985893022-70pl.png","2026-06-09T06:17:32.189729+00:00",{"id":77,"slug":78,"title":79,"cover_image":80,"image_url":80,"created_at":81,"category":13},"6f25a29c-cbb8-4f53-9af7-1656b394333a","turboquant-cuts-kv-cache-memory-6x-google-tests-zh","TurboQuant 在 Google 測試中省下 6x KV 快取","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780906682236-sqe2.png","2026-06-08T08:17:21.878314+00:00",[83,88,93,98,103,108,113,118,123,128],{"id":84,"slug":85,"title":86,"created_at":87},"f18dbadb-8c59-4723-84a4-6ad22746c77a","deepmind-bets-on-continuous-learning-ai-2026-zh","DeepMind 押注 2026 連續學習 AI","2026-03-26T08:16:02.367355+00:00",{"id":89,"slug":90,"title":91,"created_at":92},"f4a106cb-02a6-4508-8f39-9720a0a93cee","ml-papers-of-the-week-github-research-desk-zh","每週 ML 論文清單，為何紅到 GitHub","2026-03-27T01:11:39.284175+00:00",{"id":94,"slug":95,"title":96,"created_at":97},"c4f807ca-4e5f-47f1-a48c-961cf3fc44dc","ai-ml-conferences-to-watch-in-2026-zh","2026 AI 研討會投稿時程整理","2026-03-27T01:51:53.874432+00:00",{"id":99,"slug":100,"title":101,"created_at":102},"cf046742-efb2-4753-aef9-caed5da5e32e","adaptive-block-scaled-data-types-zh","IF4：神經網路量化的聰明選擇","2026-03-31T06:00:36.990273+00:00",{"id":104,"slug":105,"title":106,"created_at":107},"53a0dc54-0371-4e40-8d5e-74e94a73840c","geometry-aware-similarity-metrics-for-neural-representations-zh","超越距離測量：用微分幾何重新理解神經網路","2026-03-31T06:01:01.241968+00:00",{"id":109,"slug":110,"title":111,"created_at":112},"fee7d472-a775-4b1d-bbc2-1e8bca1bbf8b","on-the-fly-repulsion-in-the-contextual-space-for-rich-divers-zh","讓AI繪圖更有創意：用排斥力提升生成多樣性","2026-03-31T06:01:25.439673+00:00",{"id":114,"slug":115,"title":116,"created_at":117},"a9901203-d69b-447b-8854-15d14eab32b4","vision-aided-beam-prediction-cnn-eca-zh","影像輔助波束預測升級 CNN","2026-04-01T10:00:25.8073+00:00",{"id":119,"slug":120,"title":121,"created_at":122},"b55e7dd4-0a24-4b3d-804d-b0309a03f498","triple-band-fss-mimo-antenna-sub-6-ghz-zh","三頻 FSS MIMO 天線瞄準 sub-6 GHz","2026-04-01T13:18:36.857305+00:00",{"id":124,"slug":125,"title":126,"created_at":127},"f68290bd-e7f3-4b30-ba22-dcd4e0130a66","openclaw-1299-repos-eight-weeks-analysis-zh","OpenClaw 1299 個 Repo 的資料解讀","2026-04-02T05:03:45.208411+00:00",{"id":129,"slug":130,"title":131,"created_at":132},"ed9f80eb-eb02-4d35-8ad4-0ddf428751dd","beam-coherence-aware-combining-mmwave-mimo-zh","毫米波 MIMO 的雙階合併法","2026-04-02T05:27:26.897188+00:00"]