I originally posted this on github as an issue. The relay tutorial stated it would show llvm examples, but only showed an example for cuda.
I figured out how to do the llvm by digging around in other code, and I attached my modified code of the relay example to this issue report.
I think the extra examples would help anyone trying to figure out the parameters for various models, as well as how to configure for llvm and cpu context, since only the single model cuda example was provided.
I attached examples for inception v3, mobilenet, resnet, squeezenet, vgg and densenet, as well as notes about their options, all derived from digging around in the code. The densenet model may be broken, since it returned all zeros. The others returned results similar to the cuda example.