Buckets:

hf-doc-build
/

doc-dev

Files

xet

hf-doc-build/doc-dev / smol-course /pr_296 /ja /unit4 /2.html

HuggingFaceDocBuilder

about 2 months ago

download

raw

40.8 kB

	<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"自動ベンチマーク","local":"自動ベンチマーク","sections":[{"title":"自動ベンチマークの理解","local":"自動ベンチマークの理解","sections":[],"depth":2},{"title":"ベンチマークとその限界","local":"ベンチマークとその限界","sections":[{"title":"一般知識ベンチマーク","local":"一般知識ベンチマーク","sections":[],"depth":3},{"title":"推論ベンチマーク","local":"推論ベンチマーク","sections":[],"depth":3},{"title":"言語理解","local":"言語理解","sections":[],"depth":3}],"depth":2},{"title":"代替評価アプローチ","local":"代替評価アプローチ","sections":[{"title":"LLM-as-Judge","local":"llm-as-judge","sections":[],"depth":3},{"title":"評価アリーナ","local":"評価アリーナ","sections":[],"depth":3},{"title":"カスタムベンチマークスイート","local":"カスタムベンチマークスイート","sections":[],"depth":3}],"depth":2},{"title":"独自の評価戦略の作成","local":"独自の評価戦略の作成","sections":[],"depth":2},{"title":"LightEvalを使用したベンチマーク","local":"lightevalを使用したベンチマーク","sections":[{"title":"評価パイプラインの例","local":"評価パイプラインの例","sections":[],"depth":3}],"depth":2}],"depth":1}">
	<link href="/docs/smol-course/pr_296/ja/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/entry/start.afa23ac6.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/chunks/scheduler.f38f3a0e.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/chunks/singletons.6933622c.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/chunks/paths.2a0c5cb3.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/entry/app.27746c11.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/chunks/preload-helper.c5d58819.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/chunks/index.a03143fc.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/nodes/0.117e80e6.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/chunks/each.e59479a4.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/nodes/14.37c53714.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/chunks/CopyLLMTxtMenu.8710b29c.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/chunks/MermaidChart.svelte_svelte_type_style_lang.4b36369c.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/chunks/CodeBlock.a6034da5.js">
	<link rel="modulepreload" href="/docs/smol-course/pr_296/ja/_app/immutable/chunks/CourseFloatingBanner.9d00122a.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"自動ベンチマーク","local":"自動ベンチマーク","sections":[{"title":"自動ベンチマークの理解","local":"自動ベンチマークの理解","sections":[],"depth":2},{"title":"ベンチマークとその限界","local":"ベンチマークとその限界","sections":[{"title":"一般知識ベンチマーク","local":"一般知識ベンチマーク","sections":[],"depth":3},{"title":"推論ベンチマーク","local":"推論ベンチマーク","sections":[],"depth":3},{"title":"言語理解","local":"言語理解","sections":[],"depth":3}],"depth":2},{"title":"代替評価アプローチ","local":"代替評価アプローチ","sections":[{"title":"LLM-as-Judge","local":"llm-as-judge","sections":[],"depth":3},{"title":"評価アリーナ","local":"評価アリーナ","sections":[],"depth":3},{"title":"カスタムベンチマークスイート","local":"カスタムベンチマークスイート","sections":[],"depth":3}],"depth":2},{"title":"独自の評価戦略の作成","local":"独自の評価戦略の作成","sections":[],"depth":2},{"title":"LightEvalを使用したベンチマーク","local":"lightevalを使用したベンチマーク","sections":[{"title":"評価パイプラインの例","local":"評価パイプラインの例","sections":[],"depth":3}],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <div class="items-center shrink-0 min-w-[100px] max-sm:min-w-[50px] justify-end ml-auto flex" style="float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"><div class="inline-flex rounded-md max-sm:rounded-sm"><button class="inline-flex items-center gap-1 h-7 max-sm:h-7 px-2 max-sm:px-1.5 text-sm font-medium text-gray-800 border border-r-0 rounded-l-md max-sm:rounded-l-sm border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-live="polite"><span class="inline-flex items-center justify-center rounded-md p-0.5 max-sm:p-0 hover:text-gray-800 dark:hover:text-gray-200"><svg class="sm:size-3.5 size-3" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg></span> <span>Copy page</span></button> <button class="inline-flex items-center justify-center w-6 max-sm:w-5 h-7 max-sm:h-7 disabled:pointer-events-none text-sm text-gray-500 hover:text-gray-700 dark:hover:text-white rounded-r-md max-sm:rounded-r-sm border border-l transition border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-haspopup="menu" aria-expanded="false" aria-label="Open copy menu"><svg class="transition-transform text-gray-400 overflow-visible sm:size-3.5 size-3 rotate-0" width="1em" height="1em" viewBox="0 0 12 7" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M1 1L6 6L11 1" stroke="currentColor"></path></svg></button></div> </div> <h1 class="relative group"><a id="自動ベンチマーク" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#自動ベンチマーク"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>自動ベンチマーク</span></h1> <div class="flex space-x-1 absolute z-10 right-0 top-0" style=""><a href="https://discuss.huggingface.co/t/chapter-10-questions" target="_blank"><img alt="Ask a Question" class="!m-0" src="https://img.shields.io/badge/Ask%20a%20question-ffcb4c.svg?logo=data:image/svg+xml;base64,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"></a> <a href="https://colab.research.google.com/github/huggingface/smol-course/blob/main/notebooks/ja/4_evaluation/lighteval_evaluate_and_analyse_your_LLM.ipynb" target="_blank"><img alt="Open In Colab" class="!m-0" src="https://colab.research.google.com/assets/colab-badge.svg"></a> </div>

	自動ベンチマークは、さまざまなタスクや能力にわたって言語モデルを評価するための標準化されたツールとして機能します。これらはモデルの性能を理解するための有用な出発点を提供しますが、包括的な評価戦略の一部に過ぎないことを認識することが重要です。
	<h2 class="relative group"><a id="自動ベンチマークの理解" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#自動ベンチマークの理解"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>自動ベンチマークの理解</span></h2> <p data-svelte-h="svelte-8hfvzh">自動ベンチマークは通常、事前に定義されたタスクと評価指標を持つキュレーションされたデータセットで構成されます。これらのベンチマークは、基本的な言語理解から複雑な推論まで、モデルのさまざまな側面を評価することを目的としています。自動ベンチマークの主な利点はその標準化にあります。これにより、異なるモデル間で一貫した比較が可能となり、再現性のある結果が得られます。</p> <p data-svelte-h="svelte-1ldbzqb">ただし、ベンチマークの性能が必ずしも実世界での有効性に直結するわけではないことを理解することが重要です。学術的なベンチマークで優れた成績を収めたモデルでも、特定のドメインアプリケーションや実際のユースケースでは苦労することがあります。</p> <h2 class="relative group"><a id="ベンチマークとその限界" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#ベンチマークとその限界"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>ベンチマークとその限界</span></h2> <h3 class="relative group"><a id="一般知識ベンチマーク" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#一般知識ベンチマーク"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>一般知識ベンチマーク</span></h3> <p data-svelte-h="svelte-1h7u148">MMLU（Massive Multitask Language Understanding）は、科学から人文科学まで57の科目にわたる知識をテストします。包括的ではありますが、特定のドメインに必要な専門知識の深さを反映しているわけではありません。TruthfulQAは、モデルが一般的な誤解を再現する傾向を評価しますが、すべての形態の誤情報を捉えることはできません。</p> <h3 class="relative group"><a id="推論ベンチマーク" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#推論ベンチマーク"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>推論ベンチマーク</span></h3> <p data-svelte-h="svelte-gu2ali">BBH（Big Bench Hard）とGSM8Kは、複雑な推論タスクに焦点を当てています。BBHは論理的思考と計画をテストし、GSM8Kは特に数学的な問題解決を対象としています。これらのベンチマークは分析能力を評価するのに役立ちますが、実世界のシナリオで必要とされる微妙な推論を完全に捉えることはできません。</p> <h3 class="relative group"><a id="言語理解" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#言語理解"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>言語理解</span></h3> <p data-svelte-h="svelte-hk1u3h">HELMは包括的な評価フレームワークを提供し、WinoGrandeは代名詞の曖昧性解消を通じて常識をテストします。これらのベンチマークは言語処理能力に関する洞察を提供しますが、自然な会話やドメイン固有の用語の複雑さを完全に表現することはできません。</p> <h2 class="relative group"><a id="代替評価アプローチ" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#代替評価アプローチ"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>代替評価アプローチ</span></h2> <p data-svelte-h="svelte-1awprxb">多くの組織は、標準ベンチマークの限界に対処するために代替評価方法を開発しています：</p> <h3 class="relative group"><a id="llm-as-judge" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#llm-as-judge"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>LLM-as-Judge</span></h3> <p data-svelte-h="svelte-j9t3iy">ある言語モデルを使用して別のモデルの出力を評価する方法がますます人気を集めています。このアプローチは、従来の指標よりも微妙なフィードバックを提供することができますが、それ自体のバイアスと限界も伴います。</p> <h3 class="relative group"><a id="評価アリーナ" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#評価アリーナ"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>評価アリーナ</span></h3> <p data-svelte-h="svelte-v5xcn5">AnthropicのConstitutional AI Arenaのようなプラットフォームでは、モデルが相互に対話し、制御された環境で互いを評価することができます。これにより、従来のベンチマークでは明らかにならない強みと弱みが明らかになります。</p> <h3 class="relative group"><a id="カスタムベンチマークスイート" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#カスタムベンチマークスイート"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>カスタムベンチマークスイート</span></h3> <p data-svelte-h="svelte-14bee7a">組織は、特定のニーズやユースケースに合わせた内部ベンチマークスイートを開発することがよくあります。これには、ドメイン固有の知識テストや実際の展開条件を反映した評価シナリオが含まれることがあります。</p> <h2 class="relative group"><a id="独自の評価戦略の作成" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#独自の評価戦略の作成"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>独自の評価戦略の作成</span></h2> <p data-svelte-h="svelte-1qxde78">LightEvalを使用して標準ベンチマークを実行するのは簡単ですが、ユースケースに特化した評価方法の開発にも時間を投資する必要があります。</p> <p data-svelte-h="svelte-1w6vihr">標準ベンチマークは有用なベースラインを提供しますが、それだけが評価方法であってはなりません。より包括的なアプローチを開発する方法は次のとおりです：</p> <ol data-svelte-h="svelte-s1chgm"><li><p>関連する標準ベンチマークから始めて、ベースラインを確立し、他のモデルとの比較を可能にします。</p></li> <li><p>ユースケースの特定の要件と課題を特定します。モデルが実際に実行するタスクは何ですか？どのようなエラーが最も問題になりますか？</p></li> <li><p>実際のユースケースを反映したカスタム評価データセットを開発します。これには次のようなものが含まれるかもしれません：</p> <ul><li>ドメインからの実際のユーザークエリ</li> <li>遭遇した一般的なエッジケース</li> <li>特に挑戦的なシナリオの例</li></ul></li> <li><p>多層評価戦略の実装を検討します：</p> <ul><li>クイックフィードバックのための自動指標</li> <li>微妙な理解のための人間の評価</li> <li>専門家によるレビュー</li> <li>制御された環境でのA/Bテスト</li></ul></li></ol> <h2 class="relative group"><a id="lightevalを使用したベンチマーク" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#lightevalを使用したベンチマーク"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>LightEvalを使用したベンチマーク</span></h2> <p data-svelte-h="svelte-setv0j">LightEvalタスクは特定の形式を使用して定義されます：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-template-variable">{suite}</span><span class="language-xml">\|</span><span class="hljs-template-variable">{task}</span><span class="language-xml">\|</span><span class="hljs-template-variable">{num_few_shot}</span><span class="language-xml">\|</span><span class="hljs-template-variable">{auto_reduce}</span><!-- HTML_TAG_END --></pre></div> <ul data-svelte-h="svelte-kz1l"><li><strong>suite</strong>: ベンチマークスイート（例：‘mmlu’, ‘truthfulqa’）</li> <li><strong>task</strong>: スイート内の特定のタスク（例：‘abstract_algebra’）</li> <li><strong>num_few_shot</strong>: プロンプトに含める例の数（ゼロショットの場合は0）</li> <li><strong>auto_reduce</strong>: プロンプトが長すぎる場合に少数ショットの例を自動的に削減するかどうか（0または1）</li></ul> <p data-svelte-h="svelte-45tp5y">例：<code>"mmlu\|abstract_algebra\|0\|0"</code>は、ゼロショット推論でMMLUの抽象代数学タスクを評価します。</p> <h3 class="relative group"><a id="評価パイプラインの例" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#評価パイプラインの例"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>評価パイプラインの例</span></h3> <p data-svelte-h="svelte-pzo5jd">特定のドメインに関連する自動ベンチマークの評価を設定して実行する完全な例を次に示します：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> lighteval.tasks <span class="hljs-keyword">import</span> Task, Pipeline
	<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM

	<span class="hljs-comment"># 評価するタスクを定義</span>
	domain_tasks = [
	<span class="hljs-string">"mmlu\|anatomy\|0\|0"</span>,
	<span class="hljs-string">"mmlu\|high_school_biology\|0\|0"</span>,
	<span class="hljs-string">"mmlu\|high_school_chemistry\|0\|0"</span>,
	<span class="hljs-string">"mmlu\|professional_medicine\|0\|0"</span>
	]

	<span class="hljs-comment"># パイプラインパラメータを設定</span>
	pipeline_params = {
	<span class="hljs-string">"max_samples"</span>: <span class="hljs-number">40</span>, <span class="hljs-comment"># 評価するサンプル数</span>
	<span class="hljs-string">"batch_size"</span>: <span class="hljs-number">1</span>, <span class="hljs-comment"># 推論のバッチサイズ</span>
	<span class="hljs-string">"num_workers"</span>: <span class="hljs-number">4</span> <span class="hljs-comment"># ワーカープロセスの数</span>
	}

	<span class="hljs-comment"># 評価トラッカーを作成</span>
	evaluation_tracker = EvaluationTracker(
	output_path=<span class="hljs-string">"./results"</span>,
	save_generations=<span class="hljs-literal">True</span>
	)

	<span class="hljs-comment"># モデルを読み込み、パイプラインを作成</span>
	model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"your-model-name"</span>)
	pipeline = Pipeline(
	tasks=domain_tasks,
	pipeline_parameters=pipeline_params,
	evaluation_tracker=evaluation_tracker,
	model=model
	)

	<span class="hljs-comment"># 評価を実行</span>
	pipeline.evaluate()

	<span class="hljs-comment"># 結果を取得して表示</span>
	results = pipeline.get_results()
	pipeline.show_results()<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-6p5zqu">結果は次のような表形式で表示されます：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->\|<span class="hljs-string"> Task </span>\|<span class="hljs-string">Version</span>\|<span class="hljs-string">Metric</span>\|<span class="hljs-string">Value </span>\|<span class="hljs-string"> </span>\|<span class="hljs-string">Stderr</span>\|
	\|<span class="hljs-string">----------------------------------------</span>\|<span class="hljs-string">------:</span>\|<span class="hljs-string">------</span>\|<span class="hljs-string">-----:</span>\|<span class="hljs-string">---</span>\|<span class="hljs-string">-----:</span>\|
	\|<span class="hljs-string">all </span>\|<span class="hljs-string"> </span>\|<span class="hljs-string">acc </span>\|<span class="hljs-string">0.3333</span>\|<span class="hljs-string">± </span>\|<span class="hljs-string">0.1169</span>\|
	\|<span class="hljs-string">leaderboard:mmlu:_average:5 </span>\|<span class="hljs-string"> </span>\|<span class="hljs-string">acc </span>\|<span class="hljs-string">0.3400</span>\|<span class="hljs-string">± </span>\|<span class="hljs-string">0.1121</span>\|
	\|<span class="hljs-string">leaderboard:mmlu:anatomy:5 </span>\|<span class="hljs-string"> 0</span>\|<span class="hljs-string">acc </span>\|<span class="hljs-string">0.4500</span>\|<span class="hljs-string">± </span>\|<span class="hljs-string">0.1141</span>\|
	\|<span class="hljs-string">leaderboard:mmlu:high_school_biology:5 </span>\|<span class="hljs-string"> 0</span>\|<span class="hljs-string">acc </span>\|<span class="hljs-string">0.1500</span>\|<span class="hljs-string">± </span>\|<span class="hljs-string">0.0819</span>\|<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-19sk69v">結果をpandas DataFrameで処理し、視覚化や表現を自由に行うこともできます。</p> <h1 class="relative group"><a id="次のステップ" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#次のステップ"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>次のステップ</span></h1> <p data-svelte-h="svelte-y4w0jg">⏩ <a href="./custom_evaluation">カスタムドメイン評価</a>を探索し、特定のニーズに合わせた評価パイプラインの作成方法を学びましょう</p> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/smol-course/blob/main/units/ja/unit4/2.md" target="_blank"><svg class="mr-1" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M31,16l-7,7l-1.41-1.41L28.17,16l-5.58-5.59L24,9l7,7z"></path><path d="M1,16l7-7l1.41,1.41L3.83,16l5.58,5.59L8,23l-7-7z"></path><path d="M12.419,25.484L17.639,6.552l1.932,0.518L14.351,26.002z"></path></svg> <span data-svelte-h="svelte-zjs2n5"><span class="underline">Update</span> on GitHub</span></a> <p></p>

	<script>
	{
	__sveltekit_1lq7jru = {
	assets: "/docs/smol-course/pr_296/ja",
	base: "/docs/smol-course/pr_296/ja",
	env: {}
	};

	const element = document.currentScript.parentElement;

	const data = [null,null];

	Promise.all([
	import("/docs/smol-course/pr_296/ja/_app/immutable/entry/start.afa23ac6.js"),
	import("/docs/smol-course/pr_296/ja/_app/immutable/entry/app.27746c11.js")
	]).then(([kit, app]) => {
	kit.start(app, element, {
	node_ids: [0, 14],
	data,
	form: null,
	error: null
	});
	});
	}
	</script>

Xet Storage Details

Size:: 40.8 kB
Xet hash:: 4c7f82e2c228abae3c4517e9599410e8997d17ac408d9d00d20fba93ee2c93ec

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.