Has anyone here actually tried running the MRR benchmark suite from this repo yet?

[kizl01]

Curious how heavy it is to run locally and how long the 5k scenarios take in practice. Also wondering if the reference AGR and PDM implementations in the repo are close enough to the actual model internals or if they are just simplified research versions. Another thing I noticed is the iteration numbers, Chronos averages around 7.8 while baselines are 1-2 – is that really more efficient in practice or does it just mean it loops longer until it finds a fix. Anyone benchmarked it against their own models?