[[["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):2025-08-18。"],[],[],null,["# Efficient PyTorch training with cloud data\n\nVertex AI Neural Architecture Search has no requirements describing how to\ndesign your trainers. Therefore, choose any training frameworks to build the trainer.\n\nFor PyTorch training with large amounts of data, the best practice is to use the distributed training\nparadigm and to read data from Cloud Storage.\nCheck out the blog post\n[Efficient PyTorch training with Vertex AI](https://cloud.google.com/blog/products/ai-machine-learning/efficient-pytorch-training-with-vertex-ai) for methods to improve the training\nperformance. You can see an overall 6x performance improvement with data on\nCloud Storage using `WebDataset` and choosing `DistributedDataParallel` or\n`FullyShardedDataParallel` distributed training strategies. The training\nperformance using data on Cloud Storage is similar to the training performance using data on\na local disk.\n\nThe prebuilt\n[MNasNet classification example](https://github.com/google/vertex-ai-nas/blob/main/pytorch/classification/cloud_search_main.py)\nhas incorporated these methods into its training pipeline."]]