Approve and Validate Models

​

Introduction

After importing and benchmarking an AI model, the next step is deploying it to a ZEDEDA Edge Kubernetes Service cluster or cluster group and validating that it behaves as expected in production conditions. ZEDEDA Edge Inference Services supports a review and approval workflow for model deployments. A user with the Organization Admin or MLOps Admin role in the organization reviews deployment requests before they are applied to edge infrastructure. Super Admin cannot approve unless also assigned one of those roles in the organization. This ensures that only validated models reach production clusters. ZEDEDA Edge Inference Services also integrates with ZEDEDA Edge Kubernetes Service for GitOps-based deployment management. This integration tracks deployment state and approval history.

​

Prerequisites

• To submit a deployment request, you have the Organization Admin, MLOps Admin, MLOps Engineer, or Super Admin role. To approve or reject a request, you have the Organization Admin or MLOps Admin role in the organization. See Organizations, Users, and Roles.
• The model you want to deploy has been imported and is available in your model library.
• Another user with the Organization Admin or MLOps Admin role is available in your organization to review and approve deployment requests. Users cannot approve their own requests.

​

Deployment Approval Workflow

A deployment request has one of the following statuses.

​

Submit a Deployment Request

To deploy a model to a Kubernetes cluster, initiate a deployment request from the model detail page. The request enters the Pending state until an Organization Admin or MLOps Admin reviews it. For the full submission workflow — form fields, list view, status tracking, and how to update a request — see Submit Model Deployment Requests.

​

Review and Approve a Deployment (Organization Admin or MLOps Admin)

Users with the Organization Admin or MLOps Admin role review pending deployment requests and approve or reject them based on benchmark results, model validation status, and organizational policy. Users cannot approve their own requests; another user must approve.

1. Navigate to the deployment review page for your organization.
2. Select a Pending deployment request to view the model details, target infrastructure, and benchmark results.
3. Select Approve or Reject.
   • Approved deployments proceed to the target cluster or cluster group through the ZEDEDA Edge Kubernetes Service integration.
   • A rejection reason is required. Rejected deployments remain in the Rejected state with the reviewer’s reason. Rejecting a Pending Update request discards only the proposed change and returns the request to its previous status.

​

Validate a Deployed Model

After a deployment is approved and applied, validate that the model performs correctly on the target cluster or cluster group:

1. Check benchmark results from the Evaluation page to confirm the model meets your latency and throughput requirements.
2. Use Prevalidated Agents as reference implementations to compare behavior against known-good baselines.

​

In This Section

• Edge Inference Services Overview
• Inference Dashboard
• Inference Quick Start
• Access and Navigate Edge Inference Services
• Organizations, Users, and Roles
• Manage Providers
  • Manage Providers Using Jupyter Notebook Integration
• Manage Models
  • Manage Models Using Jupyter Notebook Integration
  • Benchmark Models and Manage Device Pools
  • Monitor Benchmark Runs and Interpret Results
• Submit Model Deployment Requests
• Approve and Validate Models
• Monitor Inference Jobs
• Use ZEIA to Query and Operate Your Fleet
• Use Prevalidated Agents
  • Use the Faulty Printed Circuit Board Detection Agent
  • Use the Autonomous Retail Billing Agent
• Inference Troubleshooting