This is, as of right now, a piece of garbage.
You have to hardcode your fishnet key in fishnet_benchmarker/make_benchmark.py for benchmarking to even work. Precomputed benchmark data is provided, however.
Feel free to use issues to ask for clarification about the code or point to possible errors, and to use PRs if you want to try to make this mess slightly better.
This is a script that's meant to first diagnose pricing inefficiencies in the current lichess distributed chess analysis spot instances usage, and then, if all goes well, solve them.