Jetson nano in 1.0.0-M2.1 or 1.0.0-beta7

1

In build dl4j-master.version: 1.0.0-beta7:
org.nd4j: nd4j-cuda-10.0,
org.bytedeco: cuda: 10.0-7.4-1.5,
org.deeplearning4j: deeplearning4j-core
(using ComputationGraph KerasModelImport.importKerasModelAndWeights with model, which is fine works with another card with cuda 11.6)

there is Exception:
modelimport.keras.exceptions.InvalidKerasConfigurationException: Expected model class name Model (found Functional);

In build dl4j-master.version: 1.0.0-M2.1:
org.nd4j: nd4j-cuda-10.2,
org.bytedeco: cuda: 10.2-8.2-1.5.6,
org.deeplearning4j: deeplearning4j-core

there is Exception:
Could not find jnicudart in class, module, and library paths,
Warning: Versions of org.bytedeco:javacpp:1.5.7 and org.bytedeco:cuda:10.2-8.2-1.5.6 do not match.

Does jetson nano actually work with these libraries?
Maybe these builds is missing certain dependencies?

2

@Paul don’t mix dl4j versions. That will never be supported. If you want to use the most recent M2.1 with the nano, you please use this for your backend:
https://repo1.maven.org/maven2/org/nd4j/nd4j-cuda-10.2/1.0.0-M2.1/

The standard -platform prefix doesn’t work here because it’s a different architecture. What you’ll need is to specify both nd4j-cuda-10.2 with and without the classifier as 2 separate artifacts.