你当前正在访问 Microsoft Azure Global Edition 技术文档网站。 如果需要访问由世纪互联运营的 Microsoft Azure 中国技术文档网站,请访问 https://docs.azure.cn

az ml run

Note

This reference is part of the azure-cli-ml extension for the Azure CLI (version 2.0.28 or higher). 该扩展会在首次运行 az ml run 命令时自动安装。 Learn more about extensions.

用于提交、更新和监视运行的命令。

命令

名称 说明 类型 Status
az ml run cancel

Cancel run.

Extension GA
az ml run download-logs

下载日志文件。

Extension GA
az ml run list

List runs.

Extension GA
az ml run monitor-logs

监视现有运行的日志。

Extension GA
az ml run monitor-tensorboard

使用 tensorboard 监视现有运行。

Extension GA
az ml run show

Show run.

Extension GA
az ml run submit-hyperdrive

使用运行配置提交超参数扫描。

Extension GA
az ml run submit-pipeline

从已发布的管道 ID 或管道 YAML 文件提交管道以供执行。

Extension GA
az ml run submit-script

提交脚本以供执行。

Extension GA
az ml run update

通过添加标记更新运行。

Extension GA

az ml run cancel

Cancel run.

az ml run cancel --run
                 [--experiment-name]
                 [--output-metadata-file]
                 [--path]
                 [--resource-group]
                 [--subscription-id]
                 [--workspace-name]

必需参数

--run -r

试验运行的 runId。

可选参数

The following parameters are optional, but depending on the context, one or more might become required for the command to execute successfully.

--experiment-name -e

Experiment name.

--output-metadata-file -t

为结构化对象输出提供可选的输出文件位置。

--path

用于运行配置文件的根目录的路径。

属性
默认值: .
--resource-group -g

资源组名称。

--subscription-id

指定订阅 ID。

--workspace-name -w

Workspace name.

全局参数
--debug

提高日志记录详细程度以显示所有调试日志。

属性
默认值: False
--help -h

显示此帮助消息并退出。

--only-show-errors

仅显示错误,禁止显示警告。

属性
默认值: False
--output -o

Output format.

属性
默认值: json
接受的值: json, jsonc, none, table, tsv, yaml, yamlc
--query

JMESPath 查询字符串。 有关更多信息和示例,请参阅 http://jmespath.org/

--subscription

订阅的名称或 ID。 可以使用 az account set -s NAME_OR_ID 配置默认订阅。

--verbose

增加日志记录详细程度。 使用 --debug 获取完整的调试日志。

属性
默认值: False

az ml run download-logs

下载日志文件。

az ml run download-logs --output-dir
                        --run
                        [--experiment-name]
                        [--output-metadata-file]
                        [--path]
                        [--resource-group]
                        [--subscription-id]
                        [--workspace-name]

必需参数

--output-dir -d

要下载日志文件的输出目录。

--run -r

试验运行的 runId。

可选参数

The following parameters are optional, but depending on the context, one or more might become required for the command to execute successfully.

--experiment-name -e

Experiment name.

--output-metadata-file -t

为结构化对象输出提供可选的输出文件位置。

--path

用于运行配置文件的根目录的路径。

属性
默认值: .
--resource-group -g

资源组名称。

--subscription-id

指定订阅 ID。

--workspace-name -w

Workspace name.

全局参数
--debug

提高日志记录详细程度以显示所有调试日志。

属性
默认值: False
--help -h

显示此帮助消息并退出。

--only-show-errors

仅显示错误,禁止显示警告。

属性
默认值: False
--output -o

Output format.

属性
默认值: json
接受的值: json, jsonc, none, table, tsv, yaml, yamlc
--query

JMESPath 查询字符串。 有关更多信息和示例,请参阅 http://jmespath.org/

--subscription

订阅的名称或 ID。 可以使用 az account set -s NAME_OR_ID 配置默认订阅。

--verbose

增加日志记录详细程度。 使用 --debug 获取完整的调试日志。

属性
默认值: False

az ml run list

List runs.

az ml run list [--compute-target-name]
               [--experiment-name]
               [--last]
               [--minimal]
               [--output-metadata-file]
               [--parent-run-id]
               [--path]
               [--pipeline-run-id]
               [--resource-group]
               [--status {Completed, Failed, Finalizing, Preparing, Provisioning, Queued, Running, Starting}]
               [--subscription-id]
               [--tags]
               [--workspace-name]

可选参数

The following parameters are optional, but depending on the context, one or more might become required for the command to execute successfully.

--compute-target-name

计算目标名称。

--experiment-name -e

Experiment name.

--last

提取最新的 N 元素。

属性
默认值: 10
--minimal

用于提供运行输出的最低属性的标志。

属性
默认值: False
--output-metadata-file -t

为结构化对象输出提供可选的输出文件位置。

--parent-run-id

父运行 ID。

--path

用于运行配置文件的根目录的路径。

属性
默认值: .
--pipeline-run-id

管道运行 ID。

--resource-group -g

资源组名称。

--status

运行状态。

属性
接受的值: Completed, Failed, Finalizing, Preparing, Provisioning, Queued, Running, Starting
--subscription-id

指定订阅 ID。

--tags

使用“key[=value]”语法运行的标记。

--workspace-name -w

Workspace name.

全局参数
--debug

提高日志记录详细程度以显示所有调试日志。

属性
默认值: False
--help -h

显示此帮助消息并退出。

--only-show-errors

仅显示错误,禁止显示警告。

属性
默认值: False
--output -o

Output format.

属性
默认值: json
接受的值: json, jsonc, none, table, tsv, yaml, yamlc
--query

JMESPath 查询字符串。 有关更多信息和示例,请参阅 http://jmespath.org/

--subscription

订阅的名称或 ID。 可以使用 az account set -s NAME_OR_ID 配置默认订阅。

--verbose

增加日志记录详细程度。 使用 --debug 获取完整的调试日志。

属性
默认值: False

az ml run monitor-logs

监视现有运行的日志。

az ml run monitor-logs --run
                       [--experiment-name]
                       [--output-metadata-file]
                       [--path]
                       [--resource-group]
                       [--subscription-id]
                       [--workspace-name]

必需参数

--run -r

试验运行的 runId。

可选参数

The following parameters are optional, but depending on the context, one or more might become required for the command to execute successfully.

--experiment-name -e

Experiment name.

--output-metadata-file -t

为结构化对象输出提供可选的输出文件位置。

--path

用于运行配置文件的根目录的路径。

属性
默认值: .
--resource-group -g

资源组名称。

--subscription-id

指定订阅 ID。

--workspace-name -w

Workspace name.

全局参数
--debug

提高日志记录详细程度以显示所有调试日志。

属性
默认值: False
--help -h

显示此帮助消息并退出。

--only-show-errors

仅显示错误,禁止显示警告。

属性
默认值: False
--output -o

Output format.

属性
默认值: json
接受的值: json, jsonc, none, table, tsv, yaml, yamlc
--query

JMESPath 查询字符串。 有关更多信息和示例,请参阅 http://jmespath.org/

--subscription

订阅的名称或 ID。 可以使用 az account set -s NAME_OR_ID 配置默认订阅。

--verbose

增加日志记录详细程度。 使用 --debug 获取完整的调试日志。

属性
默认值: False

az ml run monitor-tensorboard

使用 tensorboard 监视现有运行。

az ml run monitor-tensorboard --run
                              [--experiment-name]
                              [--local-directory]
                              [--output-metadata-file]
                              [--path]
                              [--resource-group]
                              [--subscription-id]
                              [--workspace-name]

必需参数

--run -r

试验运行的 runId。

可选参数

The following parameters are optional, but depending on the context, one or more might become required for the command to execute successfully.

--experiment-name -e

Experiment name.

--local-directory

用于暂存正在监视的张量板文件的本地目录。

--output-metadata-file -t

为结构化对象输出提供可选的输出文件位置。

--path

用于运行配置文件的根目录的路径。

属性
默认值: .
--resource-group -g

资源组名称。

--subscription-id

指定订阅 ID。

--workspace-name -w

Workspace name.

全局参数
--debug

提高日志记录详细程度以显示所有调试日志。

属性
默认值: False
--help -h

显示此帮助消息并退出。

--only-show-errors

仅显示错误,禁止显示警告。

属性
默认值: False
--output -o

Output format.

属性
默认值: json
接受的值: json, jsonc, none, table, tsv, yaml, yamlc
--query

JMESPath 查询字符串。 有关更多信息和示例,请参阅 http://jmespath.org/

--subscription

订阅的名称或 ID。 可以使用 az account set -s NAME_OR_ID 配置默认订阅。

--verbose

增加日志记录详细程度。 使用 --debug 获取完整的调试日志。

属性
默认值: False

az ml run show

Show run.

az ml run show --run
               [--experiment-name]
               [--output-metadata-file]
               [--path]
               [--resource-group]
               [--subscription-id]
               [--workspace-name]

必需参数

--run -r

试验运行的 runId。

可选参数

The following parameters are optional, but depending on the context, one or more might become required for the command to execute successfully.

--experiment-name -e

Experiment name.

--output-metadata-file -t

为结构化对象输出提供可选的输出文件位置。

--path

用于运行配置文件的根目录的路径。

属性
默认值: .
--resource-group -g

资源组名称。

--subscription-id

指定订阅 ID。

--workspace-name -w

Workspace name.

全局参数
--debug

提高日志记录详细程度以显示所有调试日志。

属性
默认值: False
--help -h

显示此帮助消息并退出。

--only-show-errors

仅显示错误,禁止显示警告。

属性
默认值: False
--output -o

Output format.

属性
默认值: json
接受的值: json, jsonc, none, table, tsv, yaml, yamlc
--query

JMESPath 查询字符串。 有关更多信息和示例,请参阅 http://jmespath.org/

--subscription

订阅的名称或 ID。 可以使用 az account set -s NAME_OR_ID 配置默认订阅。

--verbose

增加日志记录详细程度。 使用 --debug 获取完整的调试日志。

属性
默认值: False

az ml run submit-hyperdrive

使用运行配置提交超参数扫描。

az ml run submit-hyperdrive --hyperdrive-configuration-name
                            [--async]
                            [--conda-dependencies]
                            [--ct --target]
                            [--experiment-name]
                            [--output-metadata-file]
                            [--path]
                            [--resource-group]
                            [--run-configuration-name]
                            [--source-directory]
                            [--subscription-id]
                            [--workspace-name]
                            []

必需参数

--hyperdrive-configuration-name

hyperdrive 配置文件的全名。 该文件应位于路径参数指定的目录的子文件夹中。

可选参数

The following parameters are optional, but depending on the context, one or more might become required for the command to execute successfully.

--async

禁用输出流式处理。

属性
默认值: False
--conda-dependencies -d

重写默认的 Conda 依赖项文件。

--ct --target

要用于运行的计算目标的名称。

--experiment-name -e

Experiment name.

--output-metadata-file -t

为结构化对象输出提供可选的输出文件位置。

--path

用于运行配置文件的根目录的路径。

属性
默认值: .
--resource-group -g

资源组名称。

--run-configuration-name -c

运行配置文件的名称(无扩展名)。 该文件应位于路径参数指定的目录的子文件夹中。

--source-directory

包含源代码文件的本地目录。 如果未提供源目录,则默认为路径。

--subscription-id

指定订阅 ID。

--workspace-name -w

Workspace name.

<USER_SCRIPT_AND_ARGUMENTS>

运行提交参数,如脚本名称和脚本参数。

属性
Parameter group: Positional
全局参数
--debug

提高日志记录详细程度以显示所有调试日志。

属性
默认值: False
--help -h

显示此帮助消息并退出。

--only-show-errors

仅显示错误,禁止显示警告。

属性
默认值: False
--output -o

Output format.

属性
默认值: json
接受的值: json, jsonc, none, table, tsv, yaml, yamlc
--query

JMESPath 查询字符串。 有关更多信息和示例,请参阅 http://jmespath.org/

--subscription

订阅的名称或 ID。 可以使用 az account set -s NAME_OR_ID 配置默认订阅。

--verbose

增加日志记录详细程度。 使用 --debug 获取完整的调试日志。

属性
默认值: False

az ml run submit-pipeline

从已发布的管道 ID 或管道 YAML 文件提交管道以供执行。

az ml run submit-pipeline [--datapaths]
                          [--experiment-name]
                          [--output-metadata-file]
                          [--output_file]
                          [--parameters]
                          [--path]
                          [--pipeline-id]
                          [--pipeline-yaml]
                          [--resource-group]
                          [--subscription-id]
                          [--workspace-name]

可选参数

The following parameters are optional, but depending on the context, one or more might become required for the command to execute successfully.

--datapaths -d

用于数据路径参数分配的名称=数据存储/路径对的逗号分隔列表。

--experiment-name -n

运行提交的试验名称。 如果未指定,请使用管道名称。

--output-metadata-file -t

为结构化对象输出提供可选的输出文件位置。

--output_file -f

用于以 JSON 格式写入输出的文件。

--parameters -p

管道参数分配的名称=值对的逗号分隔列表。

--path

用于运行配置文件的根目录的路径。

属性
默认值: .
--pipeline-id -i

要提交的管道的 ID(guid)。

--pipeline-yaml -y

定义管道的 YAML 文件。

--resource-group -g

资源组名称。

--subscription-id

指定订阅 ID。

--workspace-name -w

Workspace name.

全局参数
--debug

提高日志记录详细程度以显示所有调试日志。

属性
默认值: False
--help -h

显示此帮助消息并退出。

--only-show-errors

仅显示错误,禁止显示警告。

属性
默认值: False
--output -o

Output format.

属性
默认值: json
接受的值: json, jsonc, none, table, tsv, yaml, yamlc
--query

JMESPath 查询字符串。 有关更多信息和示例,请参阅 http://jmespath.org/

--subscription

订阅的名称或 ID。 可以使用 az account set -s NAME_OR_ID 配置默认订阅。

--verbose

增加日志记录详细程度。 使用 --debug 获取完整的调试日志。

属性
默认值: False

az ml run submit-script

提交脚本以供执行。

az ml run submit-script [--async]
                        [--conda-dependencies]
                        [--ct --target]
                        [--experiment-name]
                        [--output-metadata-file]
                        [--path]
                        [--resource-group]
                        [--run-configuration-name]
                        [--source-directory]
                        [--subscription-id]
                        [--workspace-name]
                        []

可选参数

The following parameters are optional, but depending on the context, one or more might become required for the command to execute successfully.

--async

禁用输出流式处理。

属性
默认值: False
--conda-dependencies -d

重写默认的 Conda 依赖项文件。

--ct --target

要用于运行的计算目标的名称。

--experiment-name -e

Experiment name.

--output-metadata-file -t

为结构化对象输出提供可选的输出文件位置。

--path

用于运行配置文件的根目录的路径。

属性
默认值: .
--resource-group -g

资源组名称。

--run-configuration-name -c

运行配置文件的名称(无扩展名)。 该文件应位于路径参数指定的目录的子文件夹中。

--source-directory

包含源代码文件的本地目录。 如果未提供源目录,则默认为路径。

--subscription-id

指定订阅 ID。

--workspace-name -w

Workspace name.

<USER_SCRIPT_AND_ARGUMENTS>

运行提交参数,如脚本名称和脚本参数。

属性
Parameter group: Positional
全局参数
--debug

提高日志记录详细程度以显示所有调试日志。

属性
默认值: False
--help -h

显示此帮助消息并退出。

--only-show-errors

仅显示错误,禁止显示警告。

属性
默认值: False
--output -o

Output format.

属性
默认值: json
接受的值: json, jsonc, none, table, tsv, yaml, yamlc
--query

JMESPath 查询字符串。 有关更多信息和示例,请参阅 http://jmespath.org/

--subscription

订阅的名称或 ID。 可以使用 az account set -s NAME_OR_ID 配置默认订阅。

--verbose

增加日志记录详细程度。 使用 --debug 获取完整的调试日志。

属性
默认值: False

az ml run update

通过添加标记更新运行。

az ml run update --run
                 [--add-tag]
                 [--experiment-name]
                 [--output-metadata-file]
                 [--path]
                 [--resource-group]
                 [--subscription-id]
                 [--workspace-name]

必需参数

--run -r

试验运行的 runId。

可选参数

The following parameters are optional, but depending on the context, one or more might become required for the command to execute successfully.

--add-tag

使用“key[=value]”语法标记 entitiy。

--experiment-name -e

Experiment name.

--output-metadata-file -t

为结构化对象输出提供可选的输出文件位置。

--path

用于运行配置文件的根目录的路径。

属性
默认值: .
--resource-group -g

资源组名称。

--subscription-id

指定订阅 ID。

--workspace-name -w

Workspace name.

全局参数
--debug

提高日志记录详细程度以显示所有调试日志。

属性
默认值: False
--help -h

显示此帮助消息并退出。

--only-show-errors

仅显示错误,禁止显示警告。

属性
默认值: False
--output -o

Output format.

属性
默认值: json
接受的值: json, jsonc, none, table, tsv, yaml, yamlc
--query

JMESPath 查询字符串。 有关更多信息和示例,请参阅 http://jmespath.org/

--subscription

订阅的名称或 ID。 可以使用 az account set -s NAME_OR_ID 配置默认订阅。

--verbose

增加日志记录详细程度。 使用 --debug 获取完整的调试日志。

属性
默认值: False