Model warehouse, as the name implies, is a warehouse used to manage models, including importing models, viewing, adding versions, retrieving, deleting, etc. To deploy a trained model as a prediction service, it is necessary to first publish the model to a model warehouse and then publish it as a prediction service.

The model repository contains user model and common model. User models are generated by users on the platform by modeling or imported from the BOS. Common models are several sample models provided by the BML platform, which will be updated continuously later.

User Model

Select "Model warehouse" - > "User model" in the navigation column, and the page displays the list of related models created by the user.The list contains the model name, the number of versions of the model, and labels.

You can create a prediction model as follows:

1. Click "Import model" on the page
2. Select and fill in the configuration items in the pop-up window, including selecting import method (select "Import New Model" for importing a new model), filling in the model name, model version, selecting/creating model tag, selecting model type and relevant information of model file in linkage, model path and description, and then click "OK" to complete the model import.

User can also add a new version of the model based on the existing model, as follows:

1. Click "Import model" on the page
2. Select and fill in the configuration items in the pop-up window, select the import mode "import as new version of existing model", complete other configuration items and click "OK". Currently supported model types include: Deep learning type (Tensorflow-v1.13.1, paddle-fluid-v1.5.0, Pytorch-v1.1.0, Caffe2, ONNX), machine learning type (Sklearn-v0.20, GBDT-v0.82, R-v3.5.2, Pyspark-v2.4.3), General-purpose type (PMML, custom), and built-in type (transfer learning - image classification)

After importing the model, return to the model list page. Click a model name on the model list page to go to the model version list page, where you can view multiple versions of the model.

Page Operation Description:

Page [Delete] : Delete all versions of the model.

Page [Refresh]: After creating a new version, you can refresh the page to view the newly created version.

List [Delete]: Delete a specific version of the model.

List [Create Online Prediction]: Click it and fill in the configuration item in the pop-up window. After confirming, create the online prediction service with this version model.

Click a version number to enter the details page of this version, which includes model information and related prediction services. Model information includes model type, related information when creating the model, creation time, update time, model source, model tag, model path and model description. The related prediction services part displays the list of online prediction services created by this version model.
Page Operation Description:

[Create Online Prediction]: Click it and fill in the configuration item in the pop-up window. After confirming, create the online prediction service with this version model.

[Remove]: Delete this version of model

[Refresh] button: Refresh page

Edit [Model Tag]: Click the Edit button to change the tag of the model.

Edit [Model Description]: Click the Edit button to change the description of the model.

Common Module

Select "Model Repository" - > "Common Model" in the navigation bar, and the page will display the list of built-in models of the platform. Here, several sample models are included. Novice users can directly use appropriate models to establish prediction services.