在 AI Platform Training 中使用 TensorFlow 2 运行训练作业的过程与运行其他自定义代码训练作业的过程相同。但是，一些 AI Platform Training 功能在分别与 TensorFlow 2 和 TensorFlow 1 结合使用时，其工作方式存在差异。本文档对这些差异进行了总结。
[[["易于理解","easyToUnderstand","thumb-up"],["解决了我的问题","solvedMyProblem","thumb-up"],["其他","otherUp","thumb-up"]],[["很难理解","hardToUnderstand","thumb-down"],["信息或示例代码不正确","incorrectInformationOrSampleCode","thumb-down"],["没有我需要的信息/示例","missingTheInformationSamplesINeed","thumb-down"],["翻译问题","translationIssue","thumb-down"],["其他","otherDown","thumb-down"]],["最后更新时间 (UTC)：2025-04-11。"],[[["\u003cp\u003eTensorFlow 2 simplifies many APIs from TensorFlow 1 and is supported on AI Platform Training with runtime version 2.1 or later, requiring Python 3.7.\u003c/p\u003e\n"],["\u003cp\u003eDistributed training in TensorFlow 2 utilizes the \u003ccode\u003etf.distribute.Strategy\u003c/code\u003e API, with \u003ccode\u003eMultiWorkerMirroredStrategy\u003c/code\u003e or \u003ccode\u003eParameterServerStrategy\u003c/code\u003e recommended for multiple VM instances.\u003c/p\u003e\n"],["\u003cp\u003eAI Platform Training automatically sets the \u003ccode\u003eTF_CONFIG\u003c/code\u003e environment variable for each VM in TensorFlow 2, labeling the master worker as \u003ccode\u003echief\u003c/code\u003e instead of \u003ccode\u003emaster\u003c/code\u003e.\u003c/p\u003e\n"],["\u003cp\u003eTraining jobs on AI Platform can be accelerated with GPUs, using TensorFlow's \u003ccode\u003eMirroredStrategy\u003c/code\u003e for single-VM or \u003ccode\u003eMultiWorkerMirroredStrategy\u003c/code\u003e for multi-VM, or with TPUs.\u003c/p\u003e\n"],["\u003cp\u003eHyperparameter tuning with TensorFlow 2, particularly with Keras, should use \u003ccode\u003etf.summary.scalar\u003c/code\u003e to report metrics to AI Platform Training.\u003c/p\u003e\n"]]],[],null,[]]
