You can upload the configuration package into the AI Inference Server app in two different ways: Either from your local computer, including from devices and drives that your computer can access (for example, USB flash drive), or from the local Edge device on which the configuration packages can be deployed from the AI Model Manager app.
Requirements
The ZIP file of the configuration package was prepared by the AI Software Development Kit and is available in your computer or network.
Delta upload of the configuration package
The aim of the delta configuration package is to reduce the time for the pipeline import and facilitate uploading only the changes. For example, the first original package contains 600 MB of data [scripts, Python wheels], the second (delta) contains only 100 MB of data [scripts, Python wheels] and the third (delta) contains just 5 MB of data [script].
If, after a successful import procedure, there is a pipeline with the same name (pay attention to upper/lower case) but with different versions present, the AI Inference Server clones the dependencies (wheel files, scripts, configuration files) of the latest available version and combines them with the delta dependencies. The saved configuration package (for execution purposes) only contains delta dependencies (scripts and/or wheel files). The values of the variable mapping and the parameter mapping are also cloned (if possible).
The different versions of configuration packages can be handled in the same way as the basic version (first version).
If a configuration package is received via AI Model Manager and there is already a pipeline with the same package ID (UUID) and the same version AI Inference Server rejects it. (Only one configuration package with the same package ID (UUID) and the same version can be saved).
Importing the pipeline from your local computer
The welcome screen is displayed after the first login:
Follow the steps below to import the new configuration package:
Click "Overview".
Click the "Import new pipeline" button.
Select the configuration package in ZIP format.
The configuration package is imported into the AI Inference Server.
A new table is displayed on the home page:
① |
Name of the Zip file |
② |
Import progress |
③ |
Button to cancel operation |
File name ① and progress ② are displayed. You can leave the home page and switch between the tabs in this AI Inference Server, but you must not leave the user interface of the AI Interference Server altogether. The import can be canceled by clicking the button ③. The table disappears as soon as the import has finished.
If you leave the web page altogether, the import is interrupted.
The GPU package can only be received or imported in the GPU accelerated variant of the application (MLFB 6AV2170-0LA11-0AA0). The two CPU variants do not accept GPU packages.
If you try to upload a GPU pipeline package to the CPU variants, you receive an error message:
If a pipeline has already been uploaded, the AI Inference Server home screen looks like this:
If you wish to import another pipeline, click "Import new pipeline" button again and continue as before.
Pipelines can also be received from the AI Model Manager app.
Activating the pipeline
You can activate a configuration package that you previously imported or that was previously received from the AI Model Manager app. The available pipelines are shown in the section "Stored Pipelines" ①.
Select one of the available pipelines and click on the "Activate" button ②. Each pipeline can be versioned. You can get information on this in the columns "Version" and "Received on". AI Inference Server makes the selected pipeline ready to run and takes you to the pipeline visualization page where you can begin mapping variables and parameters (if needed) and run the AI model.
It is possible to activate the pipeline when a pipeline is running. You can activate one or more pipelines based on your product number, which is described in chapter Scope of this document.
You can find more information on pipeline deployment in the AI Model Manager app user manual.
Updating the model/pipeline parameter after deployment of the configuration package
It is possible to update the parameters of the model or the pipeline on the user interface of AI Inference Server and to assign some of them to a certain topic after the configuration package has been deployed. This process gives you a great deal of flexibility when making changes without transferring the pipeline to another environment and uploading the model once again to the AI Inference Server.
The parameters are displayed only when the pipeline contains such entries. The parameters must be defined by the data scientists or the author of the pipeline using the AI SDK.
Proceed as follows:
Activate a configuration package, go to the "Parameters" section.
Click "Modify" ①.
①
The "Modify" button switches to editing mode.
②
Parameter value
②
Value
③
Mapping
④
Selection of the "connection" from the drop-down menu
⑤
"Data connection setup" section
⑥
"Cancel" button
⑦
"Save" button
Select a parameter from the table.
Change the value of the parameter in the Value column ②.
If necessary, change the parameter mapping ③.
Assign a parameter to a certain topic of a pre-configured data source as input (e.g. Databus) ④.
Confirm the changes by clicking the ⑦ button.
Parameter name and parameter type cannot be changed.