airpack_scripts.airt

Tools for deploying trained neural networks on the AIR-T.

In this sense, “deploying” a neural network model means running it using a standalone inference engine (and not from within the learning framework that was used to define and train the model).

The method provided here leverages the PyCUDA interface for the shared memory buffer. PyCUDA is installed by default in AirStack, however Deepwave recommends using an Anaconda environment based on the environments/airpack_airt.yml file provide with AirPack.

Submodules

• airpack_scripts.airt.run_airt_inference