查询预制镜像仓库列表

调试

请求参数

下表仅列出该接口特有的请求参数和部分公共参数。更多信息请见公共参数。

返回参数

下表仅列出本接口特有的返回参数。更多信息请参见返回结构。

请求示例

{
	"NameContains": "pytorch"
}

返回示例

{
	"Items": [
		{
			"CreateTime": "2023-03-30T12:11:38Z",
			"Labels": [
				{
					"Key": "keyword",
					"Values": [
						"ML",
						"Training",
						"Base Image",
						"CUDA"
					]
				}
			],
			"Name": "pytorch",
			"Readme": "# 简介\n\nPytorch 镜像基于平台的 CUDA 镜像构建，新增了 `pytorch`, `torchvision`, `torchaudio` 软件包，内置 Pytorch DDP 多机测试样例。 CUDA 镜像的说明在 Pytorch 镜像中也适用。\n\n平台目前只提供 [Pytorch 官方支持的 CUDA 版本组合](https://pytorch.org/get-started/previous-versions/)。\n\n# Tag 版本列表\n\n* `2.0.0`, `2.0.0-cu11.8.0-py3.10-ubuntu20.04`\n* `1.13.1`, `1.13.1-cu11.7.0-py3.10-ubuntu20.04`\n* `1.12.1`, `1.12.1-cu11.6.0-py3.10-ubuntu20.04`\n* `1.11.0`, `1.11.0-cu11.3.0-py3.10-ubuntu20.04`\n* `1.10.1`, `1.10.1-cu11.1.1-py3.9-ubuntu20.04`\n* `2.0.0-cu11.8.0-py3.9-ubuntu20.04`\n* `2.0.0-cu11.8.0-py3.8-ubuntu20.04`\n* `2.0.0-cu11.7.0-py3.10-ubuntu20.04`\n* `2.0.0-cu11.7.0-py3.9-ubuntu20.04`\n* `2.0.0-cu11.7.0-py3.8-ubuntu20.04`\n* `1.13.1-cu11.7.0-py3.9-ubuntu20.04`\n* `1.13.1-cu11.7.0-py3.8-ubuntu20.04`\n* `1.13.1-cu11.7.0-py3.7-ubuntu20.04`\n* `1.13.1-cu11.6.0-py3.10-ubuntu20.04`\n* `1.13.1-cu11.6.0-py3.9-ubuntu20.04`\n* `1.13.1-cu11.6.0-py3.8-ubuntu20.04`\n* `1.13.1-cu11.6.0-py3.7-ubuntu20.04`\n* `1.12.1-cu11.6.0-py3.9-ubuntu20.04`\n* `1.12.1-cu11.6.0-py3.8-ubuntu20.04`\n* `1.12.1-cu11.6.0-py3.7-ubuntu20.04`\n* `1.12.1-cu11.3.0-py3.10-ubuntu20.04`\n* `1.12.1-cu11.3.0-py3.9-ubuntu20.04`\n* `1.12.1-cu11.3.0-py3.8-ubuntu20.04`\n* `1.12.1-cu11.3.0-py3.7-ubuntu20.04`\n* `1.11.0-cu11.3.0-py3.9-ubuntu20.04`\n* `1.11.0-cu11.3.0-py3.8-ubuntu20.04`\n* `1.11.0-cu11.3.0-py3.7-ubuntu20.04`\n* `1.10.1-cu11.1.1-py3.8-ubuntu20.04`\n* `1.10.1-cu11.1.1-py3.7-ubuntu20.04`\n\n# 使用方法\n\n* 直接使用：该预置镜像主要用于训练场景，创建开发机、自定义任务时选择该镜像下的任意版本均可开箱即用。\n* 作为基础镜像：若某个镜像版本的软件环境不完全满足需求时，可前往镜像仓库创建自定义镜像。构建自定义镜像的方式有如下几种：\n  * 基于现有镜像：选择该镜像版本为基础镜像，并配置上所需的 apt、pip 等软件包。\n  * 基于 Dockerfile：在编写 Dockerfile 时通过 `FROM \u003c该镜像版本的 URL\u003e` 指定该镜像版本为基础镜像，然后在 Dockerfile 中添加其它的软件包。\n  * 基于开发机：在某个使用该镜像版本的开发机中提前安装好所需软件包，并将该开发机保存成自定义镜像。\n\n## Pytorch DDP 测试\n\n镜像内置了一个基于 DDP 的 resnet50 多机训练样例，可以用来测试机器性能是否符合预期。 要进行测试，请基于该镜像提交 Pytorch DDP 框架的自定义训练任务，入口指令填写：\n\n```bash\n/opt/ml-platform/ddp.sh --num-iters 1000\n```\n\n可选的参数：\n\n```text\n  --num-iters NUM_ITERS\n                        number of benchmark iterations (default: 100)\n  --batch-size BATCH_SIZE\n                        input batch size (default: 32)\n  --num-warmup-batches NUM_WARMUP_BATCHES\n                        number of warm-up batches that don't count towards benchmark (default: 30)\n  --num-batches-per-iter NUM_BATCHES_PER_ITER\n                        number of batches per benchmark iteration (default: 10)\n  --num-classes NUM_CLASSES\n                        number of classes (default: 1000)\n```\n\n作为参考， 8 卡 A100 上平均每张卡的吞吐约为 792 images/s 。\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n",
			"UpdateTime": "2023-04-25T09:12:53Z"
		}
	],
	"PageNumber": 1,
	"PageSize": 10,
	"TotalCount": 1
}

错误码

您可访问公共错误码，获取更多错误码信息。