Compiler installation#
To deploy a custom deep learning model on Mobilint’s NPU, the model must be compiled into MXQ (Mobilint ExeCUtable) format. Mobilint’s official compiler, qb Compiler (also called qubee, qb), performs this conversion and ensures your model is optimized for execution on Mobilint hardware. It must be installed within the Docker environment provided by Mobilint.
You only need to install qb Compiler if:
You want to use your own trained model
An MXQ file is not available for the model
Tip
If you’re using precompiled models provided by Mobilint (e.g. in the Model Zoo or demo packages), you do not need to install the compiler.
Requirements#
Download the appropriate compiler package from Mobilint’s official Download Center.
System
Ubuntu 20.04.6 LTS or above
(Required only for GPU option) NVIDIA Graphics Driver 535.183.01 or above
Required package
Docker
(Required only for GPU option) nvidia-docker
See also
For more details on installing Docker, please refer to the official Docker documentations.
Installation Process#
Download the Docker image and create a container. There are two environment options: Compiling with CPU only and compiling with GPU acceleration.
Replace
{WORKING DIRECTORY}appropriately with your working directory:CPU-only option
docker pull mobilint/qbcompiler:v{QUBEE_VERSION_NUMBER}-cpu cd {WORKING DIRECTORY} docker run -it --ipc=host --name {YOUR_CONTAINER_NAME} -v $(pwd):/workspace mobilint/qbcompiler:v{QUBEE_VERSION_NUMBER}-cpu /bin/bash
With GPU option
docker pull mobilint/qbcompiler:v{QUBEE_VERSION_NUMBER} cd {WORKING DIRECTORY} docker run -it --gpus all --ipc=host --name {YOUR_CONTAINER_NAME} -v $(pwd):/workspace mobilint/qbcompiler:v{QUBEE_VERSION_NUMBER} /bin/bash
Run the following command to install
qb Compilerinto your docker container.docker cp {path/to/qubee.whl} {YOUR_CONTAINER_NAME}:/ docker start {YOUR_CONTAINER_NAME} docker exec -it {YOUR_CONTAINER_NAME} /bin/bash cd / python -m pip install qubee-{QUBEE_VERSION_NUMBER}+aries2-py3-none-any.whl