On the "Package Overview" page, select the pipeline configuration package that you want to deploy to an Edge device and click on the "Create new deployment" button to navigate to the "Create deployment process" page.
The first step of the create deployment process is to select the target device for the deployment. Activate the checkbox for the corresponding device ①. In the corresponding line, you can find the name and version of the embedding application, as well as the supported Python versions. If all details appear correct, click on the "Continue" button ②.
| ① | Checkboxes to select the devices |
| ② | "Continue" button |
| ③ | Help icon (see "Filtering deployment targets by Python version and GPU requirements" section) |
Filtering deployment targets by Python version and GPU requirements
When deploying packages, targets are filtered based on specific requirements to ensure compatibility. These restrictions affect the visibility of deployment targets in the selection table.
Python Version Compatibility
Deployment targets are filtered by the Python versions they support, as shown in the "Python version" column. If a target does not support the Python version specified in the package, it will not appear in the table.
Deploying GPU-enabled packages
The GPU-enabled packages can only be deployed to target applications that are also GPU-enabled. Therefore, the deployment is restricted to compatible targets only. On the "Create deployment process" page in the "Select target" step, the list is filtered to show only GPU-enabled target applications.
To ensure a target is visible in the table, the following conditions must be met:
the device to be online
the sufficient rights to access the device
the application to be connected to AI Asset Manager
the application running on the device must support the same Python version as in the package
the application running on the device must support the same runtime required by the package (CPU/GPU)
These restrictions help ensure that deployments are successful by matching the package requirements with the target capabilities. You can also find these requirements by hovering over the question mark icon ③ above the targets table.
Deploying to multiple devices
On the "Create deployment process" page in the "Select target" step, you can select one or more target devices. If you check the checkbox in the header of the table, all devices are selected. This allows multiple deployments to be sent to all selected devices.
After selecting the devices, click the "Continue" button. The "Deployment review" page will appear. If you've selected multiple targets, you may need to scroll to view all of them.
Note: If a warning message appears at the bottom of the "Deployment review" page, it indicates that the package includes steps using Python 3.8 or earlier. As this version will be deprecated, we recommend using newer versions of Python.
After clicking on the "Create deployment" button, the deployment process begins. The process consists of two steps:
The deployment configuration is created in AI Asset Manager with a "New" status.
The deployment action is sent to the target application in one of two ways:
a. Deployment sent through IEM: The deployment is sent via the IEM to the target, which may result in a slower process.
b. Deployment sent through Direct Connection: The deployment is sent via Direct Connection to the target, which is a faster process.
For the deployment to be sent through Direct Connection, the following requirements must be met:
The target is directly connected
The target is running a version that supports Direct Connection deployment (e.g.: AI Inference Server version 2.2.0 or higher)
Errors that may occur during the deployment sending phase are displayed in the "Deployment in progress" pop-up window.
When you navigate to the "Package Overview", you can monitor the deployment process status, which can be:
Successfully finished
In progress
Failed
Failed with redeploy (see Redeploy option section)
Redeploy option
If an error occurs during the deployment sending process and the deploy action cannot be sent to the Embedder application, there is an error message displayed for the deployment. You can try to re-send the deploy action with the existing configuration by clicking on the "Redeploy" button next to the error message.
Open target device
In the upper right corner of the pipeline, click on the menu button, to open the list of interfaces of the target device. You can open the user interface of the target device by selecting the correct IP address.
| ① | Context dropdown menu button |
| ② | Target interface list |