Tracking to Better Manage External Model Deployments | Verta.ai

Do you know where all your models are deployed?

Understanding how and where a model or model version is currently in use is fundamental to effectively managing an organization's portfolio of models throughout their lifecycle, governing models for risk and regulatory compliance, and getting the greatest value out of their data scientists' work.

Yet the very nature of how organizations frequently deploy their models across multiple networks, on edge devices or on private networks at client sites means that enterprises often lack easily accessed records of where and how a given model or version is being used.

In fact, it is not uncommon for large enterprises to have numerous models deployed in sensitive networks without any records on the purpose of these models, why they are deployed there, what tasks they perform or whether they should be replaced.

In these circumstances, it is essential that the organization put in place an effective way to track all its external deployments, whether on edge devices, air-gapped networks, or elsewhere. Tracking these deployments offers several benefits that can positively impact your organization's efficiency, accountability and security.

Enter External Deployment Tracking

Verta makes it easy to track external deployments. Verta offers both Model Catalog and Model Deployment solutions, and if you install Verta on the same network (and cluster) where you want to run most of your models, a single installation can be used to host your model catalog and Verta model endpoints. Verta automatically captures and documents Verta endpoints. (See screenshot below.)

External deployments, on other networks or devices, are right beside the Verta endpoints under the Release tab of a Registered Model Version. External endpoints can be added by selecting the “Add External Deployment” button. In addition, users who download the deployable docker image using the download icon will also be offered the quick option to log where they are deploying the model.

Users capture the necessary information using a short form, selecting if they are deploying the model on a cloud network (the most popular ones are listed) or on a private network or device (either of which can be given a name for identification purposes).

Then users provide a deployment path or location, which often is the endpoint URL on a third-party network that someone would need to access the models, assuming they have access to that network. Users also can provide a brief description and notes on the deployment.

The release page for a model lists all Verta and external deployments, which can be removed or edited from there. Additionally, teams with access to Verta's Dashboard functionality can see a count of their externally deployed versions and the distribution of models by external deployment location.

Benefits of Tracking External Deployments

Consider the following advantages that enterprises achieve by tracking their external model deployments:

• Improved Model Governance: By tracking external deployments, you gain a holistic view of how and where your models are being used. This allows you to establish better governance and ensure compliance with regulatory requirements. Documentation of the models, their purpose and their deployment locations helps maintain transparency and facilitates auditing processes.
• Enhanced Collaboration: In large enterprises, different teams or departments might deploy models independently in various networks. Tracking external deployments helps facilitate collaboration by providing a centralized view of all deployed models. This enables teams to share knowledge, avoid redundant efforts and leverage existing models to drive innovation and efficiency.
• Efficient Model Maintenance: Keeping track of external deployments enables you to proactively manage the lifecycle of your models. You can identify models that need updates, replacements or retirement based on their usage patterns and performance metrics. This ensures that your deployed models remain up-to-date, optimized and aligned with your business objectives.
• Mitigating Security Risks: Deploying models on external networks, particularly in sensitive environments, introduces security risks. By tracking external deployments, you can assess and monitor potential vulnerabilities. You can implement necessary security measures, conduct audits and ensure the models are deployed in accordance with your organization's security protocols. In the event of a security incident, tracking external deployments aids in rapid response and containment.
• Facilitating Model Reusability: External deployments might involve deploying models for specific use cases or client sites. By tracking these deployments, you can identify successful models that have the potential for reusability in other contexts. This promotes knowledge sharing, accelerates development cycles and maximizes the value derived from your deployed models.

By using a platform like Verta to track external deployments for all their models, enterprises can better manage and govern their ML assets, ultimately helping them to gain greater value from all the work that their data science teams are investing to build innovative models in the first place.

Learn more about the capabilities and benefits of Verta Model Catalog.