memory leaks - is this JVM or H2O issue?

[algomaschine]

Gents, I'm running the following code to train models from many data files in the directory. It's running for 5 min per model and like this for about 48h now. Now I notice the memory leaks are starting to appear, most of the RAM is already utilized. Seems like a garbage collector on the end of every model training is not doing its job well (because I close the h2o instance server and reinitialize it for every new model). Can you please advise what can I write in code (calling H2O methods) to perform an explicit memory cleanout after a model is saved? Thanks!

[algomaschine]

Windows Server 2016, 64bit
Python 3.7.9