[{"data":1,"prerenderedAt":205},["ShallowReactive",2],{"topic-page:\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications\u002Ffoundation-model-customization-strategies":3},{"topic":4,"category":165,"seo":169,"breadcrumbs":174,"prerequisites":184,"nextTopics":185,"relatedTopics":193},{"id":5,"slug":6,"title":7,"summary":8,"canonicalPath":9,"categoryPath":10,"difficulty":11,"targetAudience":12,"estimatedMinutes":13,"generatedAt":14,"publishedAt":14,"updatedAt":14,"reviewStatus":15,"tags":16,"expectedKnowledge":25,"learningGoal":28,"keyTerms":29,"blocks":77},"it-aws-aws-certified-ai-practitioner-foundation-model-applications-foundation-model-customization-strategies","foundation-model-customization-strategies","基盤モデルのカスタマイズ戦略","基盤モデルのカスタマイズ戦略を、AIF-C01の公式objective 3.1.5に沿って、用語、判断順序、例、誤解修正で整理します。","\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications\u002Ffoundation-model-customization-strategies","\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications","beginner","cert_candidate",6,"2026-08-06","ai_generated",[17,18,19,20,21,22,23,24],"AWS Certified AI Practitioner","AIF-C01","基盤モデルの活用","プロンプト","RAG","微調整","継続事前学習","モデル蒸留",[26,27],"AWSクラウドの基本用語を聞いたことがある","AIを使う業務例を日常的な言葉で考えられる","基盤モデルのカスタマイズ戦略について、試験で問われる違いと選択基準を説明し、短い業務例へ適用できる。",[30,35,39,46,49,52,55,62,69,73],{"id":31,"label":32,"shortDefinition":33,"role":34},"foundation-model-customization","基盤モデルのカスタマイズ","入力文脈・外部知識・追加学習などで、モデル利用の振る舞いや効率を調整することです。","primary",{"id":36,"label":37,"shortDefinition":38,"role":34},"in-context-learning","文脈内学習","モデルの重みを変えず、入力内の指示や例からその場の応答方法を示す手法です。",{"id":40,"label":41,"shortDefinition":42,"supplement":43,"role":34},"retrieval-augmented-generation","検索拡張生成（RAG）","関連情報を検索し、その結果を文脈に加えて回答を生成する方式です。",{"kind":44,"text":45},"misconception","検索結果が正しいか、回答の根拠確認は必要です。",{"id":47,"label":22,"shortDefinition":48,"role":34},"fine-tuning","事前学習済みモデルを、追加データで特定用途へ適応させる学習です。",{"id":50,"label":23,"shortDefinition":51,"role":34},"continuous-pretraining","事前学習済みモデルへ追加の大規模データを学習させ、領域知識を深める方法です。",{"id":53,"label":24,"shortDefinition":54,"role":34},"model-distillation","高性能モデルの振る舞いを、より小さなモデルへ学習させる手法です。",{"id":56,"label":57,"shortDefinition":58,"supplement":59,"role":61},"foundation-model","基盤モデル","大規模なデータで事前学習され、多様な下流タスクへ適応できるモデルです。",{"kind":44,"text":60},"すべての定型予測を、基盤モデルへ置き換える必要はありません。","contextual",{"id":63,"label":64,"shortDefinition":65,"supplement":66,"role":61},"prompt-engineering","プロンプトエンジニアリング","望む出力を得るため、指示・例・制約・形式を設計する手法です。",{"kind":67,"text":68},"comparison","コンテキスト設計は、履歴や検索結果なども含みます。",{"id":70,"label":71,"shortDefinition":72,"role":61},"pretraining","事前学習","広いデータから一般的なパターンを、利用前にモデルへ学習させる工程です。",{"id":74,"label":75,"shortDefinition":76,"role":61},"evaluation","評価","定めたデータと基準で、モデルやアプリの品質・安全性・価値を測る活動です。",[78,83,94,103,112,118,124,130,146,155],{"id":79,"type":80,"title":7,"subtitle":81,"shortDefinition":82},"hero","HeroBlock","知識更新、振る舞い変更、規模縮小を別の手段として選ぶ","目的に合わせて、入力文脈、外部知識、追加学習、または小型モデルへの知識移転を使い、モデル利用の振る舞いや効率を調整することです。",{"id":84,"type":85,"term":32,"definition":86,"plainExplanation":87,"presentation":88,"supportLabel":89,"keywords":90},"definition","DefinitionBlock","[[cloze:foundation-model-customization|基盤モデルのカスタマイズ]]とは、入力文脈、外部知識、追加学習などで振る舞いや効率を調整することです。","[[term:in-context-learning|文脈内学習]]は重みを変えず、入力内の例を使います。[[term:retrieval-augmented-generation|RAG]]は更新知識を検索し、[[term:fine-tuning|微調整]]は特定用途へ振る舞いを適応させます。[[term:continuous-pretraining|継続事前学習]]は領域知識を深め、[[term:model-distillation|モデル蒸留]]は小型モデルへ振る舞いを移します。","natural","見分けるポイント",[91,92,93],"重みを変えない方法","外部知識","追加学習",{"id":95,"type":96,"steps":97,"title":102},"decision-steps","StepBlock",[98,99,100,101],"不足が指示、最新知識、専門的振る舞い、推論効率のどれか診断する。","まず[[term:prompt-engineering|プロンプトエンジニアリング]]や少数例で要件を満たせるか試す。","更新知識ならRAG、安定した振る舞いなら微調整を比較する。","大量学習や蒸留は、長期効果が費用・データ・運用を上回る場合だけ選ぶ。","最小の変更から選ぶ",{"id":104,"type":105,"title":106,"points":107,"body":111},"exam-cues","KeyPointBlock","判断で押さえる境界",[108,109,110],"不足の種類を先に診断","知識更新はRAG、振る舞い適応は微調整が候補","重い学習ほどデータ・費用・運用責任が増える","[[term:foundation-model|基盤モデル]]へ領域知識を加える[[term:pretraining|事前学習]]は、データ・費用・運用の負担が大きくなります。改定が多い知識はRAG、安定した振る舞いは微調整を候補にし、同じ基準で[[term:evaluation|評価]]します。",{"id":113,"type":114,"src":115,"alt":116,"caption":117},"diagram","DiagramBlock","\u002Fassets\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications\u002Ffoundation-model-customization-strategies\u002Ffoundation-model-customization-strategies.svg","基盤モデルのカスタマイズ戦略についてプロンプト・例、RAG、微調整、継続事前学習・蒸留の関係を文字ラベルと矢印で示した図","変更の重さと運用負担が増える順に並べた図です。必要性を確認してから次の手段へ進みます。",{"id":119,"type":120,"scenario":121,"explanation":122,"result":123},"worked-example","ExampleBlock","毎月変わる商品在庫を答え、回答文体は一定にしたいチャットを設計します。","在庫の最新値はRAGや業務ツールから取得し、モデルの重みへ固定しません。文体は明確なプロンプトと例で評価し、不十分で大量利用の価値がある場合に微調整を比較します。最新知識と表現傾向を別問題として扱うことで、改定費用と誤情報を減らします。","一つのカスタマイズ方法へ全要件を押し込まず、役割ごとに手段を選べます。",{"id":44,"type":125,"warningTitle":126,"message":127,"severity":128,"action":129},"WarningBlock","微調整は知識データベースの代わりではない","頻繁に更新される事実を重みへ入れると、更新のたびに学習・評価が必要です。また学習した事実を正確に取り出せる保証もありません。","warning","要件を『最新事実』『出力形式』『専門振る舞い』『推論効率』へ分けます。",{"id":131,"type":132,"title":133,"questions":134},"quiz","QuizBlock","AIF-C01 判断チェック",[135],{"question":136,"choices":137,"answerIndex":141,"choiceExplanations":142},"頻繁に改定される社内規程を根拠付きで回答したい場合の第一候補はどれですか。",[138,139,140],"RAGで現行規程を検索する","改定ごとに必ず基盤モデルを事前学習する","根拠なしでモデルの記憶だけに任せる",0,[143,144,145],"正解です。更新可能な正本を回答時に参照できます。","費用と運用が過大になりやすい選択です。","改定反映と出典確認ができません。",{"id":147,"type":148,"title":149,"points":150},"summary","SummaryBlock","まとめ",[151,152,153,154],"文脈学習は重みを変えず素早く試す","RAGは更新知識を回答時に追加","微調整は振る舞い適応の候補","継続事前学習・蒸留は費用対効果を厳しく見る",{"id":156,"type":157,"title":158,"nextTopics":159,"relatedKeywords":161},"next-action","NextActionBlock","次は「AIエージェントと業務ワークフロー」へ",[160],"it-aws-aws-certified-ai-practitioner-foundation-model-applications-ai-agents-and-business-workflows",[162,163,164],"AIエージェント","業務ワークフロー","ツール",{"path":10,"title":19,"description":166,"parentPath":167,"accentToken":168},"基盤モデルの選択、RAG、プロンプト、カスタマイズ、エージェント、評価をAIF-C01の設計判断として整理します。","\u002Fit\u002Faws\u002Faws-certified-ai-practitioner","cloud-amber",{"title":170,"description":171,"canonicalUrl":172,"ogImage":173},"基盤モデルのカスタマイズ戦略 | AWS Certified AI Practitioner","AIF-C01対策として基盤モデルのカスタマイズ戦略の定義、判断軸、具体例、誤解しやすい境界を公式試験範囲に沿って学びます。","https:\u002F\u002Fzeqnilo.com\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications\u002Ffoundation-model-customization-strategies","https:\u002F\u002Fzeqnilo.com\u002Fassets\u002Fsite\u002Fdefault-og.svg",[175,178,181,182,183],{"title":176,"path":177},"IT","\u002Fit",{"title":179,"path":180},"AWS","\u002Fit\u002Faws",{"title":17,"path":167},{"title":19,"path":10},{"title":7,"path":9},[],[186],{"id":160,"title":187,"summary":188,"canonicalPath":189,"categoryPath":10,"categoryTitle":19,"difficulty":11,"targetAudience":12,"estimatedMinutes":13,"publishedAt":14,"updatedAt":14,"reviewStatus":15,"tags":190},"AIエージェントと業務ワークフロー","AIエージェントと業務ワークフローを、AIF-C01の公式objective 3.1.6に沿って、用語、判断順序、例、誤解修正で整理します。","\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications\u002Fai-agents-and-business-workflows",[17,18,19,162,163,164,191,192],"最小権限","人の承認",[194],{"id":195,"title":196,"summary":197,"canonicalPath":198,"categoryPath":10,"categoryTitle":19,"difficulty":11,"targetAudience":12,"estimatedMinutes":13,"publishedAt":14,"updatedAt":14,"reviewStatus":15,"tags":199},"it-aws-aws-certified-ai-practitioner-foundation-model-applications-rag-and-vector-databases","RAGとベクトルデータベース","RAGとベクトルデータベースを、AIF-C01の公式objective 3.1.3・3.1.4に沿って、用語、判断順序、例、誤解修正で整理します。","\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications\u002Frag-and-vector-databases",[17,18,19,21,200,201,202,203,204],"埋め込み","ベクトルデータベース","Amazon Bedrock Knowledge Bases","OpenSearch","Aurora",1787370539854]