I am not sure what is reason of using the random tuner in the autotvm tutorials? Earlier I though they don’t want to add xgboost as additional requirement to run the matmult tutorial. But this is the same case with https://github.com/apache/incubator-tvm/blob/master/tutorials/autotvm/tune_relay_x86.py which tunes the whole resnet, the tuning task is going to take hours compared to installation of xgboost module.
My understanding is that the purpose of tutorials is helping people to get start with minimum efforts (i.e., install dependencies). After you follow the tutorial and get something, you will have a sense about what you really want to achieve and make some improvements. As a result, the most important thing for tutorials is not providing an optimal script but a working flow.
For the requirements of preparing optimized configurations for end-users, CLI could be a more suitable solution. Like this discussion: [USABILITY IDEA] Create a command line compiler that would perform common build tasks out of the box