[{"data":1,"prerenderedAt":269},["ShallowReactive",2],{"docs-zh-Hant-/zh-hant/docs/service/ai/local":3},{"id":4,"title":5,"body":6,"description":261,"extension":262,"meta":263,"navigation":264,"path":265,"seo":266,"stem":267,"__hash__":268},"docs_zh_hant/zh-Hant/docs/service/ai/local.md","AI 翻譯服務（本地）",{"type":7,"value":8,"toc":250},"minimark",[9,13,17,24,28,31,35,38,41,44,47,50,53,56,59,62,65,68,72,130,133,136,144,147,153,159,162,165,168,171,177,183,186,189,192,195,201,207,210,213,235],[10,11,5],"h1",{"id":12},"ai-翻譯服務本地",[14,15,16],"p",{},"Doco Translate 支援連接本地的AI 大模型運行環境（Ollama、LM Studio、oMLX 等），進行翻譯",[14,18,19],{},[20,21],"img",{"alt":22,"src":23},"本地AI 翻譯服務","/images/docs/local-ai-services.webp",[25,26,27],"h2",{"id":27},"本地模型優勢",[14,29,30],{},"結合本地模型使用，是Doco Translate 最推薦的使用方式，他有幾個比較顯著的優勢",[32,33,34],"h4",{"id":34},"翻譯質量",[14,36,37],{},"AI 大語言模型最擅長解決自然語言的處理，尤其是翻譯場景，效果出色，翻譯品質比機器翻譯引擎更好",[32,39,40],{"id":40},"隱私安全",[14,42,43],{},"透過本地運行環境加載的模型，資料不會傳送到第三方服務商，所有處理都在本機進行，確保資料的絕對安全私密",[32,45,46],{"id":46},"成本優勢",[14,48,49],{},"比起API 方式呼叫AI 大模型服務，不需要花錢購買token 的費用，沒有額外的經濟開銷",[25,51,52],{"id":52},"本地模型劣勢",[32,54,55],{"id":55},"模型參數",[14,57,58],{},"本地模型通常是開源模型，可用於消費級設備上的模型通常參數量較小，如4B、9B、35B-A3B、27B 等參數量，相較於雲端AI 服務的模型參數量小得多，因此模型能力會遜於雲端AI 服務",[14,60,61],{},"雖然本地模型通常比雲端模型參數量小，模型能力稍弱，但翻譯本身無需依賴AI 大模型過長的思考鏈路和過強的思考能力，小參數模型在翻譯場景下完全夠用，且效果足夠好，因此，本地AI 小參數模型是最具性價比的方案",[32,63,64],{"id":64},"推理速度",[14,66,67],{},"在本機運行大模型，推理速度依賴本機的硬體效能和使用的模型參數量，機器硬體配置越好、模型參數量越小，推理速度越快，但整體速度依然比不上雲端大模型API 服務",[25,69,71],{"id":70},"本地ai-模型運行環境","本地AI 模型運行環境",[73,74,75,88],"table",{},[76,77,78],"thead",{},[79,80,81,85],"tr",{},[82,83,84],"th",{},"工具",[82,86,87],{},"特性",[89,90,91,106,118],"tbody",{},[79,92,93,103],{},[94,95,96],"td",{},[97,98,102],"a",{"href":99,"rel":100},"https://ollama.com/",[101],"nofollow","Ollama",[94,104,105],{},"使用簡單，快速部署使用模型",[79,107,108,115],{},[94,109,110],{},[97,111,114],{"href":112,"rel":113},"https://lmstudio.ai/",[101],"LM Studio",[94,116,117],{},"介面化本地模型運行環境，互動設計友好，方便使用",[79,119,120,127],{},[94,121,122],{},[97,123,126],{"href":124,"rel":125},"https://omlx.ai/",[101],"oMLX",[94,128,129],{},"專為Apple 晶片優化的本地LLM 推理環境，推理速度極快",[25,131,102],{"id":132},"ollama",[14,134,135],{},"Ollama 是上手簡單的本地模型運行工具，透過命令列操作",[14,137,138,139,143],{},"依據官網指引安裝Ollama 之後，執行 ",[140,141,142],"code",{},"ollama pull 模型名"," 下載模型",[14,145,146],{},"模型下載完成，且透過Ollama 成功運行後，即可在Doco Translate 中一鍵連接使用",[14,148,149,150],{},"Ollama 本機服務的預設請求位址為 ",[140,151,152],{},"http://localhost:11434",[14,154,155],{},[20,156],{"alt":157,"src":158},"Ollama 服務設置","/images/docs/ollama-settings.webp",[25,160,114],{"id":161},"lm-studio",[14,163,164],{},"LM Studio 是一款介面化的本機模型運行工具，上手簡單介面交互",[14,166,167],{},"在官網下載並安裝LM Studio 之後，在應用程式內下載模型",[14,169,170],{},"模型下載完成後，需要開啟LM Studio 的模型服務暴露服務接口，開啟服務後即可在Doco Translate 中一鍵連接使用",[14,172,173,174],{},"LM Studio 本機服務的預設請求位址為 ",[140,175,176],{},"http://localhost:1234",[14,178,179],{},[20,180],{"alt":181,"src":182},"LM Studio 服務設定","/images/docs/lm-studio-settings.webp",[25,184,126],{"id":185},"omlx",[14,187,188],{},"oMLX 是專為Apple 晶片優化的本地模型運行環境，模型推理速度極快",[14,190,191],{},"在官網下載安裝oMLX，安裝後進入oMLX 控制台，下載模型",[14,193,194],{},"oMLX 會自動啟動本機服務，模型下載完成後，測試可用之後即可在Doco Translate 中一鍵存取使用",[14,196,197,198],{},"oMLX 本機服務的預設請求位址為 ",[140,199,200],{},"http://localhost:8000",[14,202,203],{},[20,204],{"alt":205,"src":206},"oMLX 服務設定","/images/docs/omlx-settings.webp",[25,208,209],{"id":209},"模型推薦",[14,211,212],{},"Doco Translate 官方建議使用以下模型翻譯",[214,215,216,220,223,226,229,232],"ul",{},[217,218,219],"li",{},"Hy-MT2-1.8B：速度快，佔用記憶體極小，專為翻譯場景最佳化的模型",[217,221,222],{},"Qwen3.5:9B：速度快，佔用內容稍小",[217,224,225],{},"Qwen3.6:35B-A3B: 模型能力強，速度快，佔用記憶體稍大",[217,227,228],{},"Gemma4:12B：較為平衡，效果、速度、記憶體佔用均較為適中",[217,230,231],{},"Qwen3.6:27B: 模型能力最強，速度慢，佔用記憶體大",[217,233,234],{},"Qwen3.5:4B：速度快，佔用記憶體小，穩定性稍差",[14,236,237,238,241,242,245,246,249],{},"大多數通用情況下建議使用 ",[140,239,240],{},"Hy-MT2-1.8B"," 或 ",[140,243,244],{},"Qwen3.5:9B"," 作為主力翻譯模型，速度快記憶體佔用小，對於有著更高翻譯品質和穩定性要求的場景，建議使用 ",[140,247,248],{},"Qwen3.6:35B-A3B","，模型能力強，速度快，記憶體佔用也更大",{"title":251,"searchDepth":252,"depth":252,"links":253},"",2,[254,255,256,257,258,259,260],{"id":27,"depth":252,"text":27},{"id":52,"depth":252,"text":52},{"id":70,"depth":252,"text":71},{"id":132,"depth":252,"text":102},{"id":161,"depth":252,"text":114},{"id":185,"depth":252,"text":126},{"id":209,"depth":252,"text":209},"連接Ollama、LM Studio 或oMLX，在本機執行AI 模型進行翻譯。","md",{},true,"/zh-hant/docs/service/ai/local",{"title":5,"description":261},"zh-Hant/docs/service/ai/local","AAdr_ZVctLC1-xKO5vlhq8g8o7PQjBJ3bGNAVihB0UU",1787207635545]