This repository provides examples and explanations to help users easily get started with the Mobilint SDK qb, which includes the compiler (qbcompiler) and the runtime software (qbruntime) library.
Models converted using the compiler can be executed on the Mobilint NPU through the runtime. When properly configured, this workflow enables models to achieve faster inference performance while maintaining the original model’s accuracy.
Before getting started, ensure that you have access to a Mobilint NPU. If you don’t have one, please contact us to discuss evaluation options for your AI application.
The SDK is distributed through the Mobilint Download Center. Please sign up for an account before downloading the SDK.
Mobilint SDK qb supports the following devices:
- ARIES
- REGULUS
Both devices use qbcompiler and qbruntime, providing the same workflow for model compilation and inference. For device-specific setup, refer to the compiler and runtime tutorials.
Mobilint SDK qb consists of two main components: the compiler and the runtime.
The Mobilint qbcompiler converts models from popular deep learning frameworks into the Mobilint eXeQutable (MXQ) format. Using a pre-trained model and a calibration dataset, the compiler parses, quantizes, and optimizes the model for execution on the Mobilint NPU.
The Mobilint qbruntime enables execution of the compiled MXQ models on the NPU. Using the runtime library, you can integrate your compiled MXQ models into real-world applications in a simple and efficient way.
For more information, please refer to the compiler and runtime tutorials.
compilation/- Tutorials for compiling models using the Mobilint qbcompilerruntime/- Tutorials for running compiled models using the Mobilint qbruntimeruntime/python/- Python runtime tutorialsruntime/cpp/- C++ cross-compiled runtime tutorials
assets/- Images and diagrams used in the documentation
If you encounter any issues while following this tutorial, please contact our technical support email.