i read the doc from page NVDLA Primer,In-memory data formats, it says “NVDLA has a mechanism to reduce memory bandwidth by sparsely storing convolution weights”,“NVDLA engine support weight sparse compression option”, is the function opened by default or how to enable it? and i want to check is it works, how can i do ? please help me, thanks very much
Thank you for the question.
We are checking this with our internal team and will share more information with you later.
The feature is enabled in the DLA 1.3.1, which is not available for Jetson yet.
Please wait for our future release for the feature.
thanks for you reply! addition,“The feature is enabled in the DLA 1.3.1”, how to check the DLA version or the current feature it support? is there any release note or document or pags?
In JetPack4.5, the DLA version is 1.3.0.
Currently, you will need to use TensorRT API for DLA.
So the support matrix can be found in the TensorRT document: