[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-how-to-run-and-fine-tune-llms-with-unsloth-zh":3,"article-related-how-to-run-and-fine-tune-llms-with-unsloth-zh":30,"series-tools-6a68ec02-df15-4ef4-8cdd-7fbcf23d2f3b":79},{"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},"6a68ec02-df15-4ef4-8cdd-7fbcf23d2f3b","how-to-run-and-fine-tune-llms-with-unsloth-zh","怎麼用 Unsloth 跑與微調 LLM","\u003Cp data-speakable=\"summary\">這篇教你用 Unsloth 在\u003Ca href=\"\u002Fnews\u002Fopenhuman-private-personal-ai-local-setup-zh\">本機\u003C\u002Fa>載入 \u003Ca href=\"\u002Ftag\u002Fllm\">LLM\u003C\u002Fa>，完成 LoRA 微調前置設定，並跑通一次最小訓練驗證。\u003C\u002Fp>\u003Cp>這篇給想把\u003Ca href=\"\u002Fnews\u002Fmistral-ai-models-ranked-2026-zh\">模型\u003C\u002Fa>下載、推理、微調串成一條本機流程的開發者看。照做完，你會拿到一套可重複使用的 Unsloth 環境、可成功載入的模型，以及一個能直接延伸到正式訓練的工作流。\u003C\u002Fp>\u003Cp>如果你常在不同模型家族之間切換，這份指南也會幫你先確認環境、再選對教程、最後用一次短訓練檢查把風險壓到最低。\u003C\u002Fp>\u003Ch2>開始之前\u003C\u002Fh2>\u003Cul>\u003Cli>Python 3.10 或 Python 3.11\u003C\u002Fli>\u003Cli>pip 24+\u003C\u002Fli>\u003Cli>Node 不需要；需要的是可用的 NVIDIA GPU 與對應 CUDA 驅動，或 Unsloth 支援的本機執行環境\u003C\u002Fli>\u003Cli>Hugging Face 帳號\u003C\u002Fli>\u003Cli>Hugging Face access token\u003C\u002Fli>\u003Cli>Git 已安裝\u003C\u002Fli>\u003Cli>系統記憶體至少 16 GB\u003C\u002Fli>\u003Cli>GPU 記憶體足以容納你要載入的模型\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>Step 1: 建立 Unsloth 虛擬環境\u003C\u002Fh2>\u003Cp>目的：先隔離套件版本，避免 Python 依賴互相污染，讓後續載入模型時更穩定。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779671157661-sl8g.png\" alt=\"怎麼用 Unsloth 跑與微調 LLM\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cpre>\u003Ccode>python -m venv .venv\nsource .venv\u002Fbin\u002Factivate  # Windows: .venv\\Scripts\\activate\npip install --upgrade pip\npip install unsloth\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>驗收：你應該看到 Unsloth 安裝完成，且執行 \u003Ccode>python -c \"import unsloth; print('ok')\"\u003C\u002Fcode> 會輸出 \u003Ccode>ok\u003C\u002Fcode>。\u003C\u002Fp>\u003Ch2>Step 2: 登入 Hugging Face\u003C\u002Fh2>\u003Cp>目的：先完成授權，才能下載受限模型與 Unsloth 教程中引用的 checkpoint。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779671168362-x2fl.png\" alt=\"怎麼用 Unsloth 跑與微調 LLM\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cpre>\u003Ccode>huggingface-cli login\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>驗收：你應該看到登入成功的提示，並且本機已保存 \u003Ca href=\"\u002Ftag\u002Ftoken\">token\u003C\u002Fa>，可用來後續拉取模型。\u003C\u002Fp>\u003Ch2>Step 3: 打開對應模型教程\u003C\u002Fh2>\u003Cp>目的：先選對模型家族，再決定推理或微調路線，避免把不同 checkpoint 的設定混在一起。\u003C\u002Fp>\u003Cp>先從 \u003Ca href=\"https:\u002F\u002Funsloth.ai\u002Fdocs\u002Fmodels\u002Ftutorials\" target=\"_blank\" rel=\"noopener noreferrer\">Unsloth LLM Tutorials\u003C\u002Fa> 入口開始，再打開你要用的模型家族頁面；原始碼與範例也可參考 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Funslothai\u002Funsloth\" target=\"_blank\" rel=\"noopener noreferrer\">Unsloth GitHub repo\u003C\u002Fa>。每個教程都會標明模型名稱、載入方式與適合的微調步驟。\u003C\u002Fp>\u003Cp>驗收：你應該看到的是具體模型頁面，而不是泛用首頁，且頁面名稱要和你準備使用的 checkpoint 家族一致。\u003C\u002Fp>\u003Ch2>Step 4: 載入本機推理模型\u003C\u002Fh2>\u003Cp>目的：先確認模型能在你的機器上成功下載與初始化，再進入訓練，避免把時間花在錯誤環境上。\u003C\u002Fp>\u003Cpre>\u003Ccode>from unsloth import FastLanguageModel\n\nmodel, tokenizer = FastLanguageModel.from_pretrained(\n    model_name = \"unsloth\u002FQwen3-8B\",\n    max_seq_length = 2048,\n    load_in_4bit = True,\n)\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>驗收：你應該看到模型與 tokenizer 成功載入，且過程沒有 out-of-memory 錯誤。若載入失敗，先改小 checkpoint，再調低 \u003Ccode>max_seq_length\u003C\u002Fcode>。\u003C\u002Fp>\u003Ch2>Step 5: 轉成可訓練的 LoRA 模型\u003C\u002Fh2>\u003Cp>目的：把已載入的基礎模型包成可微調狀態，讓你能用較少參數完成適配。\u003C\u002Fp>\u003Cpre>\u003Ccode>model = FastLanguageModel.get_peft_model(\n    model,\n    r = 16,\n    target_modules = [\"q_proj\", \"k_proj\", \"v_proj\", \"o_proj\"],\n    lora_alpha = 16,\n    lora_dropout = 0,\n    bias = \"none\",\n    use_gradient_checkpointing = \"unsloth\",\n    random_state = 3407,\n)\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>驗收：你應該看到 LoRA adapters 已掛到模型上，而且可訓練參數數量會明顯小於原始模型。\u003C\u002Fp>\u003Ch2>Step 6: 跑一次短訓練檢查\u003C\u002Fh2>\u003Cp>目的：在正式訓練前先做最小可行驗證，確認資料管線、訓練迴圈與儲存流程都正常。\u003C\u002Fp>\u003Cpre>\u003Ccode># 簡化的 smoke test\n# trainer.train()\n# trainer.save_model(\".\u002Funsloth-checkpoint\")\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>驗收：你應該看到至少一個訓練步驟完成，並且產生一個 checkpoint 資料夾。若失敗，先檢查 batch size、context length 與 \u003Ca href=\"\u002Ftag\u002Fgpu\">GPU\u003C\u002Fa> 記憶體。\u003C\u002Fp>\u003Ch2>常見錯誤\u003C\u002Fh2>\u003Cul>\u003Cli>模型太大，GPU 記憶體不夠。修法：改用較小 checkpoint、開啟 4-bit 載入，或降低 \u003Ccode>max_seq_length\u003C\u002Fcode>。\u003C\u002Fli>\u003Cli>還沒登入 Hugging Face 就開始下載。修法：先執行 \u003Ccode>huggingface-cli login\u003C\u002Fcode>，再確認 token 已保存。\u003C\u002Fli>\u003Cli>把不同模型家族的教程混著用。修法：回到 Unsloth 教程索引，改跟同一個 checkpoint 家族的頁面走。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>接下來可以看什麼\u003C\u002Fh2>\u003Cp>當本機載入與短訓練都成功後，下一步就去看對應\u003Ca href=\"\u002Fnews\u002Fwhy-washington-is-underreacting-to-ai-security-models-zh\">模型的\u003C\u002Fa>正式 fine-tuning 教程，再接著處理 chat template、資料格式與部署流程，讓這套 Unsloth 工作流可以直接進入應用環境。\u003C\u002Fp>","這篇教你用 Unsloth 先完成本機 LLM 載入，再做 LoRA 微調與最小可行訓練驗證。","unsloth.ai","https:\u002F\u002Funsloth.ai\u002Fdocs\u002Fmodels\u002Ftutorials",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779671157661-sl8g.png","tools","zh","3ef54137-04c2-4eee-9e36-794f5d769b6b",[17,18,19,20,21],"Unsloth","LoRA","Hugging Face","Python","CUDA",[23,24,25],"先用虛擬環境與 Hugging Face 登入把基礎設好，再開始載入模型。","先完成本機推理驗證，再把模型轉成 LoRA 可訓練狀態。","用短訓練 smoke test 先檢查資料、記憶體與儲存流程是否正常。",3,"2026-05-25T01:05:27.703266+00:00","2026-05-25T01:05:27.587+00:00","be8b3ef0-c6e7-477d-a0e9-f8f1e74e0335",{"tags":31,"relatedLang":11,"relatedPosts":42},[32,34,36,38,40],{"name":17,"slug":33},"unsloth",{"name":20,"slug":35},"python",{"name":18,"slug":37},"lora",{"name":19,"slug":39},"hugging-face",{"name":21,"slug":41},"cuda",[43,49,55,61,67,73],{"id":44,"slug":45,"title":46,"cover_image":47,"image_url":47,"created_at":48,"category":13},"5656a6ab-9e07-41be-9cea-3440fb8846e2","nvidia-lg-ai-collaboration-playbook-zh","Nvidia 和 LG 把 AI 合作變成模板","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781056994999-8eng.png","2026-06-10T02:02:46.590133+00:00",{"id":50,"slug":51,"title":52,"cover_image":53,"image_url":53,"created_at":54,"category":13},"e48be66d-d7de-419e-b5fd-805f0784ef15","ollama-best-free-ai-path-2026-zh","Ollama 是 2026 年真正適合工作的免費 AI 路徑","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781056077878-11pc.png","2026-06-10T01:47:24.632993+00:00",{"id":56,"slug":57,"title":58,"cover_image":59,"image_url":59,"created_at":60,"category":13},"9b53427c-8c2a-4960-a773-f14d4528caae","awesome-production-ml-turns-chaos-into-stack-zh","這份 MLOps 清單把混亂拆成堆疊","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781055220958-dmar.png","2026-06-10T01:33:14.850634+00:00",{"id":62,"slug":63,"title":64,"cover_image":65,"image_url":65,"created_at":66,"category":13},"d5af1522-28aa-4cfb-8779-1ecf168bc0b5","bentoml-turns-model-serving-into-python-apis-zh","BentoML 把模型服務變成 Python API","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781054310299-c1gm.png","2026-06-10T01:17:56.193093+00:00",{"id":68,"slug":69,"title":70,"cover_image":71,"image_url":71,"created_at":72,"category":13},"63d8b456-ad6b-475e-86e9-d4677ca226aa","magenta-realtime-2-score-inside-daw-zh","Magenta RealTime 2 讓你在 DAW 裡即時改曲","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781046204038-8tox.png","2026-06-09T23:02:55.9651+00:00",{"id":74,"slug":75,"title":76,"cover_image":77,"image_url":77,"created_at":78,"category":13},"f60261ff-a42e-4cfb-9f90-97785e633289","open-source-ai-tools-beat-claude-paid-tiers-zh","開源 AI 工具在價值上已經贏過 Claude 付費方案","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781045266035-on7t.png","2026-06-09T22:47:20.195939+00:00",[80,85,90,95,100,105,110,115,120,125],{"id":81,"slug":82,"title":83,"created_at":84},"855cd52f-6fab-46cc-a7c1-42195e8a0de4","surepath-real-time-mcp-policy-controls-zh","SurePath 推出即時 MCP 政策控管","2026-03-26T07:57:40.77233+00:00",{"id":86,"slug":87,"title":88,"created_at":89},"9b19ab54-edef-4dbd-9ce4-a51e4bae4ebb","mcp-in-2026-the-ai-tool-layer-teams-use-zh","2026 年 MCP：團隊真的在用的 AI 工具層","2026-03-26T08:01:46.589694+00:00",{"id":91,"slug":92,"title":93,"created_at":94},"af9c46c3-7a28-410b-9f04-32b3de30a68c","prompting-in-2026-what-actually-works-zh","2026 提示工程，真正有用的是什麼","2026-03-26T08:08:12.453028+00:00",{"id":96,"slug":97,"title":98,"created_at":99},"05553086-6ed0-4758-81fd-6cab24b575e0","garry-tan-open-sources-claude-code-toolkit-zh","Garry Tan 開源 Claude Code 工具包","2026-03-26T08:26:20.068737+00:00",{"id":101,"slug":102,"title":103,"created_at":104},"042a73a2-18a2-433d-9e8f-9802b9559aac","github-ai-projects-to-watch-in-2026-zh","2026 必看 20 個 GitHub AI 專案","2026-03-26T08:28:09.619964+00:00",{"id":106,"slug":107,"title":108,"created_at":109},"a5f94120-ac0d-4483-9a8b-63590071ac6a","claude-code-vs-cursor-2026-zh","Claude Code 與 Cursor 深度對比：202…","2026-03-26T13:27:14.279193+00:00",{"id":111,"slug":112,"title":113,"created_at":114},"0975afa1-e0c7-4130-a20d-d890eaed995e","practical-github-guide-learning-ml-2026-zh","2026 機器學習入門 GitHub 實用指南","2026-03-27T01:16:49.712576+00:00",{"id":116,"slug":117,"title":118,"created_at":119},"bfdb467a-290f-4a80-b3a9-6f081afb6dff","aiml-2026-student-ai-ml-lab-repo-review-zh","AIML-2026：像課綱的學生實驗 Repo","2026-03-27T01:21:51.467798+00:00",{"id":121,"slug":122,"title":123,"created_at":124},"80cabc3e-09fc-4ff5-8f07-b8d68f5ae545","ai-trending-github-repos-and-research-feeds-zh","AI Trending：把 AI 資源收成一張表","2026-03-27T01:31:35.262183+00:00",{"id":126,"slug":127,"title":128,"created_at":129},"3ce6e6e2-bac5-463e-9f8d-45caabcc61f7","awesome-ai-for-science-research-tools-map-zh","AI 科研工具清單，開始像地圖了","2026-03-27T01:46:50.521945+00:00"]