Cloud Compiler
Port and optimize your custom AI models for various hardware platforms using DeGirum’s Cloud Compiler.
Last updated
Was this helpful?
Port and optimize your custom AI models for various hardware platforms using DeGirum’s Cloud Compiler.
Last updated
Was this helpful?
The DeGirum Cloud Compiler simplifies the process of preparing AI models for real-world deployment. It takes PyTorch checkpoints—specifically models trained using the Ultralytics repository—and compiles them into formats that run efficiently on supported edge AI hardware.
Upload your PyTorch checkpoint to the Cloud Compiler and let it handle the conversion. You can adjust parameters to optimize performance, choose target runtimes and devices, and provide a custom dataset for quantization to meet your deployment requirements.
When a model is compiled, it can be loaded with PySDK or run directly in the browser.
Access to the Cloud Compiler is on a per-workspace basis.
When your workspace has access to the Cloud Compiler, you will be able to access the Cloud Compiler. Without access, you can request early access for your workspace to the Cloud Compiler through the form available in the Cloud Compiler section of the AI Hub.
Use the AI Hub account selector to select a Workspace with Cloud Compiler access
Click your account name in the top right corner of the AI Hub and select a workspace with Cloud Compiler access.
Create or get access to a Model Zoo
The Cloud Compiler places compiled models into a model zoo. Ensure you have a model zoo where you can upload pre-compiled models.
To learn how to create Model Zoos, .
Visit the Cloud Compiler page
After ensuring you have access to a Model Zoo where you can upload models, click Cloud Compiler in the AI Hub navigation bar to visit the Cloud Compiler page.
Upload a checkpoint file
Click Upload File to submit a PyTorch checkpoint (.pt) for compilation.
Fill out the Details section
In the Details section, fill in details such as name prefix, version, image width, and image height to identify your model in the model zoo.
Select a Model Zoo in the Details section
Choose the Model Zoo where the compiled model will be published.
If the Model Zoo selector is empty, then recheck if you have created or can access a Model Zoo where you can upload pre-compiled mode.
Select a runtime and device in the Target section
In the Target section, choose the target runtime, device, and type from the dropdown menus. The Cloud Compiler uses your selections to build your model.
Select advanced options (optional)
Each runtime and device offers advanced options. View the advanced options to further optimize your model, such as by uploading a calibration dataset for quantization.
Start model compilation
Click Compile to start. The Cloud Compiler task will appear in the task list available when you click Tasks in the AI Hub navigation bar. When the process completes, your model is published to the selected model zoo, and you will receive an email confirming that the model has been compiled.