커스텀 학습 작업, 하이퍼파라미터 조정 작업 또는 학습 파이프라인을 Vertex AI에 제출하기 전에 Python 학습 애플리케이션 또는 커스텀 컨테이너를 만들어 Vertex AI에서 실행하려는 학습 코드 및 종속 항목을 정의해야 합니다. TensorFlow, PyTorch, scikit-learn 또는 XGBoost를 사용하여 Python 학습 애플리케이션을 만드는 경우에는 사전 빌드된 컨테이너를 사용하여 코드를 실행할 수 있습니다. 어떤 옵션을 선택해야 할지 잘 모르겠다면 학습 코드 요구사항을 참조하세요.
분산형 학습
여러 작업자 풀을 지정하여 커스텀 학습 작업, 하이퍼파라미터 조정 작업 또는 분산 학습을 위한 학습 파이프라인을 구성할 수 있습니다.
첫 번째 작업자 풀을 사용하여 기본 복제본을 구성하고 복제본 카운트를 1로 설정합니다.
머신러닝 프레임워크가 분산형 학습에 이러한 추가 클러스터 작업을 지원하는 경우 작업자 복제본, 파라미터 서버 복제본 또는 평가자 복제본을 구성하는 작업자 풀을 추가합니다.
[[["이해하기 쉬움","easyToUnderstand","thumb-up"],["문제가 해결됨","solvedMyProblem","thumb-up"],["기타","otherUp","thumb-up"]],[["이해하기 어려움","hardToUnderstand","thumb-down"],["잘못된 정보 또는 샘플 코드","incorrectInformationOrSampleCode","thumb-down"],["필요한 정보/샘플이 없음","missingTheInformationSamplesINeed","thumb-down"],["번역 문제","translationIssue","thumb-down"],["기타","otherDown","thumb-down"]],["최종 업데이트: 2025-09-04(UTC)"],[],[],null,["# Choose a custom training method\n\nIf you're writing your own training code instead of using AutoML},\nthere are\nseveral ways of doing custom training to consider. This document provides a\nbrief overview and comparison of the different ways you can run custom training.\n\nCustom training resources on Vertex AI\n--------------------------------------\n\nThere are three types of Vertex AI resources you can create to\ntrain custom models on Vertex AI:\n\n- [Custom jobs](/vertex-ai/docs/training/create-custom-job)\n- [Hyperparameter tuning jobs](/vertex-ai/docs/training/using-hyperparameter-tuning)\n- [Training pipelines](/vertex-ai/docs/training/create-training-pipeline)\n\nWhen you create a *custom job*, you specify settings that Vertex AI\nneeds to run your training code, including:\n\n- One for single-node training ([`WorkerPoolSpec`](/vertex-ai/docs/reference/rest/v1/CustomJobSpec#workerpoolspec)), or multiple worker pools for distributed training\n- Optional settings for configuring job scheduling ([`Scheduling`](/vertex-ai/docs/reference/rest/v1/CustomJobSpec#scheduling)), [setting\n certain environment variables for your training\n code](/vertex-ai/docs/training/code-requirements#environment-variables), [using a custom\n service account](/vertex-ai/docs/general/custom-service-account), and [using VPC Network\n Peering](/vertex-ai/docs/general/vpc-peering)\n\nWithin the worker pool(s), you can specify the following settings:\n\n- [Machine types and accelerators](/vertex-ai/docs/training/configure-compute)\n- [Configuration of what type of training code the worker pool\n runs](/vertex-ai/docs/training/configure-container-settings): either a Python training application ([`PythonPackageSpec`](/vertex-ai/docs/reference/rest/v1/CustomJobSpec#pythonpackagespec)) or a custom container ([`ContainerSpec`](/vertex-ai/docs/reference/rest/v1/CustomJobSpec#containerspec))\n\n[Hyperparameter tuning jobs](/vertex-ai/docs/training/using-hyperparameter-tuning) have additional settings to configure, such as the\nmetric. Learn more about\n[hyperparameter tuning](/vertex-ai/docs/training/hyperparameter-tuning-overview).\n\nA *training pipeline* orchestrates custom training jobs or hyperparameter tuning\njobs with additional steps, such as loading a dataset or uploading the model to\nVertex AI after the training job is successfully completed.\n\n### Custom training resources\n\nTo view existing training pipelines in your project, go to the **Training\nPipelines** page in the **Vertex AI** section of the\nGoogle Cloud console.\n\n[Go to Training pipelines](https://console.cloud.google.com/vertex-ai/training/training-pipelines)\n| **Note:** The **Training pipelines** page shows AutoML training pipelines, in addition to custom training pipelines. You can use the **Model type** column to distinguish between the two.\n\nTo view existing custom jobs in your project, go to the **Custom jobs** page.\n\n[Go to Custom jobs](https://console.cloud.google.com/vertex-ai/training/custom-jobs)\n\nTo view existing hyperparameter tuning jobs in your project, go to the\n**Hyperparameter tuning** page.\n\n[Go to Hyperparameter tuning](https://console.cloud.google.com/vertex-ai/training/hyperparameter-tuning-jobs)\n\nPrebuilt and custom containers\n------------------------------\n\nBefore you submit a custom training job, hyperparameter tuning job, or a\ntraining pipeline to Vertex AI, you need to create a [Python\ntraining application](/vertex-ai/docs/training/create-python-pre-built-container) or a [custom container](/vertex-ai/docs/training/create-custom-container) to define the training code and\ndependencies you want to run on Vertex AI. If you create a Python\ntraining application using TensorFlow, PyTorch, scikit-learn, or XGBoost, you\ncan use our prebuilt containers to run your code. If you're not sure which of\nthese options to choose, refer to the\n[training code requirements](/vertex-ai/docs/training/code-requirements) to learn more.\n\nDistributed training\n--------------------\n\nYou can configure a custom training job, hyperparameter tuning job, or a\ntraining pipeline for distributed training by specifying multiple\nworker pools:\n\n- Use your first worker pool to configure your primary replica, and set the replica count to 1.\n- Add more worker pools to configure worker replicas, parameter server replicas, or evaluator replicas, if your machine learning framework supports these additional cluster tasks for distributed training.\n\nLearn more about\n[using distributed training](/vertex-ai/docs/training/distributed-training).\n\nWhat's next\n-----------\n\n- Learn how to [create a persistent resource](/vertex-ai/docs/training/persistent-resource-create) to run custom training jobs.\n- See [Create custom training jobs](/vertex-ai/docs/training/create-custom-job) to learn how to create custom training jobs to run your custom training applications on Vertex AI.\n- See [Create training pipelines](/vertex-ai/docs/training/create-training-pipeline) to learn how to create training pipelines to run custom training applications on Vertex AI.\n- See [Use hyperparameter tuning](/vertex-ai/docs/training/using-hyperparameter-tuning) to learn about Hyperparameter tuning searches."]]