This article mainly introduces how to convert neural network models of different platforms into model codes and data that can be run on the NPU.
The documentation is adapted to the NPU driver version 126.96.36.199. For drivers and demo repositories and SDKs lower than this version, the documentation is also applicable, but not completely consistent.
$ mkdir workspace && cd workspace
Enter the SDK directory,
The SDK is mainly divided into several
conversion tools and
compilation tools, and
acuity-toolkit #Conversion tool , used to convert AI models
Since all linux codes have been supports local compiled, host compilation is no longer supported. Therefore, the contents of linux_sdk have been completely removed.
Enter the Docs directory,
DDK_Application_Guide_0.7.pdf describes each document. The document records a series of processes from conversion to integration, as well as some common problems
acuity-toolkit is the conversion tool directory,
The main directory of interest is
1. bin #Conversion is a collection of various tools used, most of which are not open source.
The environment dependency package required by the conversion tool can be installed directly on the PC or installed through the virtual environment
tensorflow==2.0.0 can be replaced by
The conversion script is in the
Use scripts to convert AI models
After the conversion is completed, you can see the converted code in the
xxxx_nbg_unify directory, here is the built-in model as an example
For the setting of conversion parameters, please refer to ‘Model_Transcoding and Running User Guide Eng.pdf’ in