Buckets:

hf-doc-build
/

doc-dev

Files

xet

hf-doc-build/doc-dev / transformers /main /zh /hpo_train.html

rtrm

about 1 month ago

download

raw

30.8 kB

	<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"使用Trainer API进行超参数搜索","local":"使用trainer-api进行超参数搜索","sections":[{"title":"超参数搜索后端","local":"超参数搜索后端","sections":[],"depth":2},{"title":"如何在示例中启用超参数搜索","local":"如何在示例中启用超参数搜索","sections":[],"depth":2},{"title":"针对DDP微调的超参数搜索","local":"针对ddp微调的超参数搜索","sections":[],"depth":2}],"depth":1}">
	<link href="/docs/transformers/main/zh/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/entry/start.a61b9c50.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/scheduler.9991993c.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/singletons.2822fe91.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/index.02cfeb18.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/paths.d66588b4.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/entry/app.99775688.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/index.7fc9a5e7.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/nodes/0.f4c5a5c1.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/each.e59479a4.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/nodes/13.ec04f431.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/CodeBlock.e11cba92.js">
	<link rel="modulepreload" href="/docs/transformers/main/zh/_app/immutable/chunks/EditOnGithub.84ab7f0e.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"使用Trainer API进行超参数搜索","local":"使用trainer-api进行超参数搜索","sections":[{"title":"超参数搜索后端","local":"超参数搜索后端","sections":[],"depth":2},{"title":"如何在示例中启用超参数搜索","local":"如何在示例中启用超参数搜索","sections":[],"depth":2},{"title":"针对DDP微调的超参数搜索","local":"针对ddp微调的超参数搜索","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="使用trainer-api进行超参数搜索" 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="#使用trainer-api进行超参数搜索"><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>使用Trainer API进行超参数搜索</span></h1> <p data-svelte-h="svelte-8j0u42">🤗 Transformers库提供了一个优化过的<code>Trainer</code>类，用于训练🤗 Transformers模型，相比于手动编写自己的训练循环，这更容易开始训练。<code>Trainer</code>提供了超参数搜索的API。本文档展示了如何在示例中启用它。</p> <h2 class="relative group"><a id="超参数搜索后端" 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="#超参数搜索后端"><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>超参数搜索后端</span></h2> <p data-svelte-h="svelte-as9xxk"><code>Trainer</code> 目前支持四种超参数搜索后端：<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></p> <p data-svelte-h="svelte-1owi0b4">在使用它们之前，您应该先安装它们作为超参数搜索后端。</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="如何在示例中启用超参数搜索" 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="#如何在示例中启用超参数搜索"><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>如何在示例中启用超参数搜索</span></h2> <p data-svelte-h="svelte-3ulmkt">定义超参数搜索空间，不同的后端需要不同的格式。</p> <p data-svelte-h="svelte-s7uas8">对于sigopt，请参阅sigopt <a href="https://docs.sigopt.com/ai-module-api-references/api_reference/objects/object_parameter" rel="nofollow">object_parameter</a>，它类似于以下内容：</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">>>> </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">"bounds"</span>: {<span class="hljs-string">"min"</span>: <span class="hljs-number">1e-6</span>, <span class="hljs-string">"max"</span>: <span class="hljs-number">1e-4</span>}, <span class="hljs-string">"name"</span>: <span class="hljs-string">"learning_rate"</span>, <span class="hljs-string">"type"</span>: <span class="hljs-string">"double"</span>},
	<span class="hljs-meta">... </span> {
	<span class="hljs-meta">... </span> <span class="hljs-string">"categorical_values"</span>: [<span class="hljs-string">"16"</span>, <span class="hljs-string">"32"</span>, <span class="hljs-string">"64"</span>, <span class="hljs-string">"128"</span>],
	<span class="hljs-meta">... </span> <span class="hljs-string">"name"</span>: <span class="hljs-string">"per_device_train_batch_size"</span>,
	<span class="hljs-meta">... </span> <span class="hljs-string">"type"</span>: <span class="hljs-string">"categorical"</span>,
	<span class="hljs-meta">... </span> },
	<span class="hljs-meta">... </span> ]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1q93awm">对于optuna，请参阅optuna <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>，它类似于以下内容：</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">>>> </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">"learning_rate"</span>: trial.suggest_float(<span class="hljs-string">"learning_rate"</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">"per_device_train_batch_size"</span>: trial.suggest_categorical(<span class="hljs-string">"per_device_train_batch_size"</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-1ocmljl">Optuna提供了多目标HPO。您可以在<code>hyperparameter_search</code>中传递<code>direction</code>参数，并定义自己的<code>compute_objective</code>以返回多个目标值。在<code>hyperparameter_search</code>中将返回Pareto Front（<code>List[BestRun]</code>），您应该参考<a href="https://github.com/huggingface/transformers/blob/main/tests/trainer/test_trainer.py" rel="nofollow">test_trainer</a>中的测试用例<code>TrainerHyperParameterMultiObjectOptunaIntegrationTest</code>。它类似于以下内容：</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">>>> </span>best_trials = trainer.hyperparameter_search(
	<span class="hljs-meta">... </span> direction=[<span class="hljs-string">"minimize"</span>, <span class="hljs-string">"maximize"</span>],
	<span class="hljs-meta">... </span> backend=<span class="hljs-string">"optuna"</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> <p data-svelte-h="svelte-1w57ahn">对于raytune，可以参考raytune的<a href="https://docs.ray.io/en/latest/tune/api/search_space.html" rel="nofollow">object_parameter</a>，它类似于以下内容：</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">>>> </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">"learning_rate"</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">"per_device_train_batch_size"</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-j3ax4r">对于wandb，可以参考wandb的<a href="https://docs.wandb.ai/guides/sweeps/configuration" rel="nofollow">object_parameter</a>，它类似于以下内容：</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">>>> </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">"method"</span>: <span class="hljs-string">"random"</span>,
	<span class="hljs-meta">... </span> <span class="hljs-string">"metric"</span>: {<span class="hljs-string">"name"</span>: <span class="hljs-string">"objective"</span>, <span class="hljs-string">"goal"</span>: <span class="hljs-string">"minimize"</span>},
	<span class="hljs-meta">... </span> <span class="hljs-string">"parameters"</span>: {
	<span class="hljs-meta">... </span> <span class="hljs-string">"learning_rate"</span>: {<span class="hljs-string">"distribution"</span>: <span class="hljs-string">"uniform"</span>, <span class="hljs-string">"min"</span>: <span class="hljs-number">1e-6</span>, <span class="hljs-string">"max"</span>: <span class="hljs-number">1e-4</span>},
	<span class="hljs-meta">... </span> <span class="hljs-string">"per_device_train_batch_size"</span>: {<span class="hljs-string">"values"</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-1ik81ii">定义一个<code>model_init</code>函数并将其传递给[Trainer]，作为示例：</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">>>> </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">".ckpt"</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> use_auth_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-w9fjru">使用你的<code>model_init</code>函数、训练参数、训练和测试数据集以及评估函数创建一个<code>Trainer</code>。</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">>>> </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-1fogicn">调用超参数搜索，获取最佳试验参数，后端可以是<code>"optuna"</code>/<code>"sigopt"</code>/<code>"wandb"</code>/<code>"ray"</code>。方向可以是<code>"minimize"</code>或<code>"maximize"</code>，表示是否优化更大或更低的目标。</p> <p data-svelte-h="svelte-1k3ogdl">您可以定义自己的compute_objective函数，如果没有定义，将调用默认的compute_objective，并将评估指标（如f1）之和作为目标值返回。</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">>>> </span>best_trial = trainer.hyperparameter_search(
	<span class="hljs-meta">... </span> direction=<span class="hljs-string">"maximize"</span>,
	<span class="hljs-meta">... </span> backend=<span class="hljs-string">"optuna"</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="针对ddp微调的超参数搜索" 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="#针对ddp微调的超参数搜索"><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>针对DDP微调的超参数搜索</span></h2> <p data-svelte-h="svelte-wf0c68">目前，Optuna和Sigopt已启用针对DDP的超参数搜索。只有rank-zero进程会进行超参数搜索并将参数传递给其他进程。</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/zh/hpo_train.md" target="_blank"><span data-svelte-h="svelte-1kd6by1"><</span> <span data-svelte-h="svelte-x0xyl0">></span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a> <p></p>

	<script>
	{
	__sveltekit_173uja2 = {
	assets: "/docs/transformers/main/zh",
	base: "/docs/transformers/main/zh",
	env: {}
	};

	const element = document.currentScript.parentElement;

	const data = [null,null];

	Promise.all([
	import("/docs/transformers/main/zh/_app/immutable/entry/start.a61b9c50.js"),
	import("/docs/transformers/main/zh/_app/immutable/entry/app.99775688.js")
	]).then(([kit, app]) => {
	kit.start(app, element, {
	node_ids: [0, 13],
	data,
	form: null,
	error: null
	});
	});
	}
	</script>

Xet Storage Details

Size:: 30.8 kB
Xet hash:: 58dc175335ea20710456297a5997a4b7210bf78a974371d69031c2fbdcbbc637

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