Buckets:

rtrm's picture
download
raw
29.6 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Trainer API를 사용한 하이퍼파라미터 탐색&quot;,&quot;local&quot;:&quot;hyperparameter-search-using-trainer-api&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;하이퍼파라미터 탐색 백엔드&quot;,&quot;local&quot;:&quot;hyperparameter-search-backend&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;예제에서 하이퍼파라미터 탐색을 활성화하는 방법&quot;,&quot;local&quot;:&quot;how-to-enable-hyperparameter-search-in-example&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;DDP 미세 조정을 위한 하이퍼파라미터 탐색&quot;,&quot;local&quot;:&quot;hyperparameter-search-for-ddp-finetune&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}">
<link href="/docs/transformers/main/ko/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/entry/start.9aa88961.js">
<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/scheduler.9bc65507.js">
<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/singletons.9eec45c3.js">
<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/index.3b203c72.js">
<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/paths.566078f7.js">
<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/entry/app.84fb67c3.js">
<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/index.707bf1b6.js">
<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/nodes/0.1c99376b.js">
<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/each.e59479a4.js">
<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/nodes/20.86a089fc.js">
<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/CodeBlock.54a9f38d.js">
<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/EditOnGithub.922df6ba.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Trainer API를 사용한 하이퍼파라미터 탐색&quot;,&quot;local&quot;:&quot;hyperparameter-search-using-trainer-api&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;하이퍼파라미터 탐색 백엔드&quot;,&quot;local&quot;:&quot;hyperparameter-search-backend&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;예제에서 하이퍼파라미터 탐색을 활성화하는 방법&quot;,&quot;local&quot;:&quot;how-to-enable-hyperparameter-search-in-example&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;DDP 미세 조정을 위한 하이퍼파라미터 탐색&quot;,&quot;local&quot;:&quot;hyperparameter-search-for-ddp-finetune&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="hyperparameter-search-using-trainer-api" 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="#hyperparameter-search-using-trainer-api"><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>Trainer API를 사용한 하이퍼파라미터 탐색</span></h1> <p data-svelte-h="svelte-1wpadea">🤗 Transformers에서는 🤗 Transformers 모델을 학습시키는데 최적화된 <code>Trainer</code> 클래스를 제공하기 때문에, 사용자는 직접 훈련 루프를 작성할 필요 없이 더욱 간편하게 학습을 시킬 수 있습니다. 또한, <code>Trainer</code>는 하이퍼파라미터 탐색을 위한 API를 제공합니다. 이 문서에서 이 API를 활용하는 방법을 예시와 함께 보여드리겠습니다.</p> <h2 class="relative group"><a id="hyperparameter-search-backend" 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="#hyperparameter-search-backend"><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>하이퍼파라미터 탐색 백엔드</span></h2> <p data-svelte-h="svelte-14yr5bp"><code>Trainer</code>는 현재 아래 4가지 하이퍼파라미터 탐색 백엔드를 지원합니다:
<a href="https://optuna.org/" rel="nofollow">optuna</a><a href="https://sigopt.com/" rel="nofollow">sigopt</a>, <a href="https://docs.ray.io/en/latest/tune/index.html" rel="nofollow">raytune</a>, <a href="https://wandb.ai/site/sweeps" rel="nofollow">wandb</a> 입니다.</p> <p data-svelte-h="svelte-1n3obee">하이퍼파라미터 탐색 백엔드로 사용하기 전에 아래의 명령어를 사용하여 라이브러리들을 설치하세요.</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 -->pip install optuna/sigopt/wandb/ray[tune] <!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="how-to-enable-hyperparameter-search-in-example" 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="#how-to-enable-hyperparameter-search-in-example"><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>예제에서 하이퍼파라미터 탐색을 활성화하는 방법</span></h2> <p data-svelte-h="svelte-1aiezuq">하이퍼파라미터 탐색 공간을 정의하세요. 하이퍼파라미터 탐색 백엔드마다 서로 다른 형식이 필요합니다.</p> <p data-svelte-h="svelte-wifdo4">sigopt의 경우, 해당 <a href="https://docs.sigopt.com/ai-module-api-references/api_reference/objects/object_parameter" rel="nofollow">object_parameter</a> 문서를 참조하여 아래와 같이 작성하세요:</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-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">sigopt_hp_space</span>(<span class="hljs-params">trial</span>):
<span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> [
<span class="hljs-meta">... </span> {<span class="hljs-string">&quot;bounds&quot;</span>: {<span class="hljs-string">&quot;min&quot;</span>: <span class="hljs-number">1e-6</span>, <span class="hljs-string">&quot;max&quot;</span>: <span class="hljs-number">1e-4</span>}, <span class="hljs-string">&quot;name&quot;</span>: <span class="hljs-string">&quot;learning_rate&quot;</span>, <span class="hljs-string">&quot;type&quot;</span>: <span class="hljs-string">&quot;double&quot;</span>},
<span class="hljs-meta">... </span> {
<span class="hljs-meta">... </span> <span class="hljs-string">&quot;categorical_values&quot;</span>: [<span class="hljs-string">&quot;16&quot;</span>, <span class="hljs-string">&quot;32&quot;</span>, <span class="hljs-string">&quot;64&quot;</span>, <span class="hljs-string">&quot;128&quot;</span>],
<span class="hljs-meta">... </span> <span class="hljs-string">&quot;name&quot;</span>: <span class="hljs-string">&quot;per_device_train_batch_size&quot;</span>,
<span class="hljs-meta">... </span> <span class="hljs-string">&quot;type&quot;</span>: <span class="hljs-string">&quot;categorical&quot;</span>,
<span class="hljs-meta">... </span> },
<span class="hljs-meta">... </span> ]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-15tcz6x">optuna의 경우, 해당 <a href="https://optuna.readthedocs.io/en/stable/tutorial/10_key_features/002_configurations.html#sphx-glr-tutorial-10-key-features-002-configurations-py" rel="nofollow">object_parameter</a> 문서를 참조하여 아래와 같이 작성하세요:</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-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">optuna_hp_space</span>(<span class="hljs-params">trial</span>):
<span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> {
<span class="hljs-meta">... </span> <span class="hljs-string">&quot;learning_rate&quot;</span>: trial.suggest_float(<span class="hljs-string">&quot;learning_rate&quot;</span>, <span class="hljs-number">1e-6</span>, <span class="hljs-number">1e-4</span>, log=<span class="hljs-literal">True</span>),
<span class="hljs-meta">... </span> <span class="hljs-string">&quot;per_device_train_batch_size&quot;</span>: trial.suggest_categorical(<span class="hljs-string">&quot;per_device_train_batch_size&quot;</span>, [<span class="hljs-number">16</span>, <span class="hljs-number">32</span>, <span class="hljs-number">64</span>, <span class="hljs-number">128</span>]),
<span class="hljs-meta">... </span> }<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1q8ydvq">raytune의 경우, 해당 <a href="https://docs.ray.io/en/latest/tune/api/search_space.html" rel="nofollow">object_parameter</a> 문서를 참조하여 아래와 같이 작성하세요:</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-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">ray_hp_space</span>(<span class="hljs-params">trial</span>):
<span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> {
<span class="hljs-meta">... </span> <span class="hljs-string">&quot;learning_rate&quot;</span>: tune.loguniform(<span class="hljs-number">1e-6</span>, <span class="hljs-number">1e-4</span>),
<span class="hljs-meta">... </span> <span class="hljs-string">&quot;per_device_train_batch_size&quot;</span>: tune.choice([<span class="hljs-number">16</span>, <span class="hljs-number">32</span>, <span class="hljs-number">64</span>, <span class="hljs-number">128</span>]),
<span class="hljs-meta">... </span> }<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1a9peeq">wandb의 경우, 해당 <a href="https://docs.wandb.ai/guides/sweeps/configuration" rel="nofollow">object_parameter</a> 문서를 참조하여 아래와 같이 작성하세요:</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-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">wandb_hp_space</span>(<span class="hljs-params">trial</span>):
<span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> {
<span class="hljs-meta">... </span> <span class="hljs-string">&quot;method&quot;</span>: <span class="hljs-string">&quot;random&quot;</span>,
<span class="hljs-meta">... </span> <span class="hljs-string">&quot;metric&quot;</span>: {<span class="hljs-string">&quot;name&quot;</span>: <span class="hljs-string">&quot;objective&quot;</span>, <span class="hljs-string">&quot;goal&quot;</span>: <span class="hljs-string">&quot;minimize&quot;</span>},
<span class="hljs-meta">... </span> <span class="hljs-string">&quot;parameters&quot;</span>: {
<span class="hljs-meta">... </span> <span class="hljs-string">&quot;learning_rate&quot;</span>: {<span class="hljs-string">&quot;distribution&quot;</span>: <span class="hljs-string">&quot;uniform&quot;</span>, <span class="hljs-string">&quot;min&quot;</span>: <span class="hljs-number">1e-6</span>, <span class="hljs-string">&quot;max&quot;</span>: <span class="hljs-number">1e-4</span>},
<span class="hljs-meta">... </span> <span class="hljs-string">&quot;per_device_train_batch_size&quot;</span>: {<span class="hljs-string">&quot;values&quot;</span>: [<span class="hljs-number">16</span>, <span class="hljs-number">32</span>, <span class="hljs-number">64</span>, <span class="hljs-number">128</span>]},
<span class="hljs-meta">... </span> },
<span class="hljs-meta">... </span> }<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1q7kvtf"><code>model_init</code> 함수를 정의하고 이를 <code>Trainer</code>에 전달하세요. 아래는 그 예시입니다.</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-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">model_init</span>(<span class="hljs-params">trial</span>):
<span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> AutoModelForSequenceClassification.from_pretrained(
<span class="hljs-meta">... </span> model_args.model_name_or_path,
<span class="hljs-meta">... </span> from_tf=<span class="hljs-built_in">bool</span>(<span class="hljs-string">&quot;.ckpt&quot;</span> <span class="hljs-keyword">in</span> model_args.model_name_or_path),
<span class="hljs-meta">... </span> config=config,
<span class="hljs-meta">... </span> cache_dir=model_args.cache_dir,
<span class="hljs-meta">... </span> revision=model_args.model_revision,
<span class="hljs-meta">... </span> token=<span class="hljs-literal">True</span> <span class="hljs-keyword">if</span> model_args.use_auth_token <span class="hljs-keyword">else</span> <span class="hljs-literal">None</span>,
<span class="hljs-meta">... </span> )<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-17wxp1q">아래와 같이 <code>model_init</code> 함수, 훈련 인수, 훈련 및 테스트 데이터셋, 그리고 평가 함수를 사용하여 <code>Trainer</code>를 생성하세요:</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-meta">&gt;&gt;&gt; </span>trainer = Trainer(
<span class="hljs-meta">... </span> model=<span class="hljs-literal">None</span>,
<span class="hljs-meta">... </span> args=training_args,
<span class="hljs-meta">... </span> train_dataset=small_train_dataset,
<span class="hljs-meta">... </span> eval_dataset=small_eval_dataset,
<span class="hljs-meta">... </span> compute_metrics=compute_metrics,
<span class="hljs-meta">... </span> tokenizer=tokenizer,
<span class="hljs-meta">... </span> model_init=model_init,
<span class="hljs-meta">... </span> data_collator=data_collator,
<span class="hljs-meta">... </span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-ikq9mj">하이퍼파라미터 탐색을 호출하고, 최적의 시험 매개변수를 가져오세요. 백엔드는 <code>&quot;optuna&quot;</code>/<code>&quot;sigopt&quot;</code>/<code>&quot;wandb&quot;</code>/<code>&quot;ray&quot;</code> 중에서 선택할 수 있습니다. 방향은 <code>&quot;minimize&quot;</code> 또는 <code>&quot;maximize&quot;</code> 중 선택하며, 목표를 최소화할 것인지 최대화할 것인지를 결정합니다.</p> <p data-svelte-h="svelte-1k38imx">자신만의 compute_objective 함수를 정의할 수 있습니다. 만약 이 함수를 정의하지 않으면, 기본 compute_objective가 호출되고, f1과 같은 평가 지표의 합이 목푯값으로 반환됩니다.</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-meta">&gt;&gt;&gt; </span>best_trial = trainer.hyperparameter_search(
<span class="hljs-meta">... </span> direction=<span class="hljs-string">&quot;maximize&quot;</span>,
<span class="hljs-meta">... </span> backend=<span class="hljs-string">&quot;optuna&quot;</span>,
<span class="hljs-meta">... </span> hp_space=optuna_hp_space,
<span class="hljs-meta">... </span> n_trials=<span class="hljs-number">20</span>,
<span class="hljs-meta">... </span> compute_objective=compute_objective,
<span class="hljs-meta">... </span>)<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="hyperparameter-search-for-ddp-finetune" 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="#hyperparameter-search-for-ddp-finetune"><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>DDP 미세 조정을 위한 하이퍼파라미터 탐색</span></h2> <p data-svelte-h="svelte-13bo63v">현재, DDP(Distributed Data Parallelism; 분산 데이터 병렬처리)를 위한 하이퍼파라미터 탐색은 optuna와 sigopt에서 가능합니다. 최상위 프로세스가 하이퍼파라미터 탐색 과정을 시작하고 그 결과를 다른 프로세스에 전달합니다.</p> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/transformers/blob/main/docs/source/ko/hpo_train.md" target="_blank"><span data-svelte-h="svelte-1kd6by1">&lt;</span> <span data-svelte-h="svelte-x0xyl0">&gt;</span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a> <p></p>
<script>
{
__sveltekit_1hrx8 = {
assets: "/docs/transformers/main/ko",
base: "/docs/transformers/main/ko",
env: {}
};
const element = document.currentScript.parentElement;
const data = [null,null];
Promise.all([
import("/docs/transformers/main/ko/_app/immutable/entry/start.9aa88961.js"),
import("/docs/transformers/main/ko/_app/immutable/entry/app.84fb67c3.js")
]).then(([kit, app]) => {
kit.start(app, element, {
node_ids: [0, 20],
data,
form: null,
error: null
});
});
}
</script>

Xet Storage Details

Size:
29.6 kB
·
Xet hash:
5b0a052c1f0f3e367699664be2a691d692bee94f6292b8f7c53ab1d59dc8d5be

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