Unfortunately the benchmark tests script is not being actively maintained on the reference page you have linked to. I’ll add to our backlog to review it, but to enable you to run your tests independent of that, below are some remarks.
To load all CPU cores you can run stress:
stress --cpu 4
If you are using other HW units than CPU, you may wan to stress them as well together with the CPU, as it will generate more heat.
For GPU, one could use glmark. You can build it using OpenEmbedded. See the references:
For IPU and VPU, if it makes sense that you also need to account for heating generated by them, you may try to come up with a HW accelerated Gstreamer pipeline to e.g. encode video from a webcam and/or play an x264 econded video in a loop. See references:
p.s. in case you haven’t yet come across it, there are some references about thermal management and temperature monitoring that may be useful: