At dAIEDGE we have developed our dAIEDGE-VLab, a virtual environment that allows testing, comparing and optimising AI models on different hardware platforms in a secure, flexible and collaborative way.
This virtual lab makes it easy for researchers and developers to deploy models on real or simulated devices, generating performance metrics that allow them to make informed decisions about their viability in real-world scenarios.
- dAIEDGE-VLab Architecture
- Web interface: At the top level of its architecture, dAIEdge-VLab offers a benchmarking solution to end users via a web interface through which they can create benchmarks.
- Target machine and host: At the bottom level of the architecture, hardware boards, called targets, are configured to enable benchmarking of the model.
- Target repository: The scripts for executing the VLab benchmarking steps are not stored directly on the host computer, but in a separate GitLab repository, called the "target repository". When running a benchmark, the target repository is cloned on the host computer and the scripts are executed.
- dAIEdge-VLab CI/CD pipeline: In the middle tier of the architecture, when running a performance test with a specific configuration (model, inference engine, target), the dAIEdge-VLab CI/CD pipeline is triggered with the corresponding variables/tags (model, inference engine, target).
- Gitlab executor and Docker container: To link the target-specific host machines to the VLab server, the Gitlab executor is used. For each target, an executor is registered on the VLab server with specific tags (hardware board, inference engine). The registered executor must be active and running on its specific host machine.
In this initial phase, access will be restricted to partners, relevant organisations, and researchers, including participants in the open call, as necessary.