Export a model using the Export General-Purpose Model component-Platform For AI(PAI)-阿里云帮助中心

The Export General-Purpose Model component saves a trained MaxCompute model to Object Storage Service (OSS). Use it as the final step in a PAI Designer pipeline after training — the exported model is then available for deployment or sharing.

Prerequisites

Before you begin, ensure that you have:

Completed OSS authorization for Designer. For details, see Cloud product dependencies and authorization: Designer

Supported upstream components

The following training components can connect to the Export General-Purpose Model component. Port requirements and supported export formats differ by component type.

Upstream component	Type	Supported formats	Port connection
Gradient Boosting Decision Trees (GBDT) Binary Classification	General ML	original, pmml	Port 1 only
Linear Support Vector Machine	General ML	original, pmml	Port 1 only
Logistic Regression Binary Classification	General ML	original, pmml	Port 1 only
GBDT Regression	General ML	original, pmml	Port 1 only
Linear Regression	General ML	original, pmml	Port 1 only
k-means Clustering	General ML	original, pmml	Port 1 only
Logistic Regression Multiclass Classification	General ML	original, pmml	Port 1 only
Naive Bayes	General ML	original, pmml	Port 1 only
PS-SMART Binary Classification Training	PS-series	original only	Port 1 and port 2
PS-SMART Multiclass Classification	PS-series	original only	Port 1 and port 2
PS-SMART Regression	PS-series	original only	Port 1 and port 2
PS Linear Regression	PS-series	original only	Port 1 and port 2

Port connection rules:

General ML components — connect the output model to port 1 of the Export General-Purpose Model component.
PS-series components — connect to both port 1 and port 2.

Configure the component in Designer

On the Parameters tab of the Export General-Purpose Model component, configure the following parameters.

Parameter	Required	Description	Default
Rename	No	A new name for the exported model. If left blank, the model is saved using the name of the upstream output model.	None (model keeps upstream name)
Overwrite Existing Model	No	When used together with Rename, overwrites the previously exported model on each run. Configure this to avoid accumulating multiple exports at the same OSS path.	Not selected
Overwrite	No	Overwrites any model file of the same name at the specified OSS path.	Not selected
Export Model Format	Yes	The format of the exported model file. Supported values: original (exports a `.meta` file and a `.model` file) or pmml (exports a `.xml` file). Check the table above to confirm which formats your upstream component supports.	None

To overwrite the previously exported model on every run, set Rename and select Overwrite Existing Model.

Configure the component using PAI commands

Run the following PAI command in an SQL script or an ODPS SQL node in DataWorks.

PAI -name generalmodeltransfer2oss
-project algo_public
-Dformat="original"
-Drename="model_export"
-Doverwrite="false"
-DossPath="oss://examplebucket-cn-hangzhou-internal.aliyuncs.com/export/"
-Darn="acs:ram::xxxxxxxxxx:role/aliyunodpspaidefaultrole"
-DmodelName="model_flow_aius5tamq5rv4x****_node_anprs9ufo40opc****_model"

Parameter	Required	Description	Default
`format`	Yes	Export format: `original` or `pmml`.	None
`rename`	No	New name for the exported model.	None (model keeps upstream name)
`overwrite`	No	Whether to overwrite a model of the same name at the target path. Accepted values: `true` or `false`.	`false`
`ossPath`	Yes	The OSS path to export the model to.	None
`arn`	Yes	The authorization information (ARN of the RAM role). To get this value, go to the PAI console, navigate to All Product Dependencies, find the Designer section, and click View Authorization Information in the Actions column. For details, see Cloud product dependencies and authorization: Designer.	None
`inputTable`	No	Required when exporting from a PS-series component. Specify the model table generated by the training component.	None
`modelName`	Yes	The name of the offline model to export.	None