Build Process Overview

The EdgeBench platform offers a streamlined, step-by-step workflow for deploying AI models to edge hardware. The journey begins with the User, who acts as the central driver of the process—creating projects, uploading models, configuring builds, and generating results.

This section outlines the step-by-step process for deploying AI models using the platform. The process consists of these key steps:

1. Start with a Project

The user initiates the workflow by creating a Project, which serves as a container for organizing related models and builds. Each project is defined by its name, task type, and framework.

2. Choose or Import a Model

The user can either select a pre-trained model from the ADI Model Zoo or upload a custom model using the BYOM feature. Uploaded models must follow specific format and packaging requirements and are stored within the Workspace.

3. Select Target Device

With the model in place, the user chooses the Device on which the model will be deployed. This selection determines hardware-specific compilation settings.

4. Configure and Compile

The user sets Compilation Flags and Optimization Levels to tailor the build for performance, power efficiency, or compatibility. These settings are saved as part of the Configuration, which is used to generate a Build—a deployable unit combining the model and device settings.

5. Upload Input Data

The user provides Input Files for inference, which may include images, audio, or time-series data. These files are used to test the model's performance on real-world data.

6. Add Pre-Processing Script (Required for BYOM)

To ensure compatibility between input data and model expectations, the user uploads a Pre-Processing Script. This script transforms raw input into a format suitable for inference.

7. Add Post-Processing Script (Required for BYOM)

The user defines how the model's output should be interpreted using a Post-Processing Script, which formats results for analysis or downstream use.

8. Run Inference

With all components in place, the user initiates Inference on the selected device. A progress monitor provides real-time feedback during execution.

9. View Results

Once inference is complete, the user accesses the Results, which include output data, performance metrics, and accuracy scores. These results are timestamped and organized by batch for easy tracking.

Throughout the journey, a fixed navigation panel remains visible, highlighting the current step and helping users stay oriented as they move through the workflow.

The detailed setup instruction for each step follows in the succeeding sections.