How to train and use my own neural network model for ltex?

[Saltsmart]

Hello! I have seen these options: ltex.additionalRules.languageModel, ltex.additionalRules.neuralNetworkModel and ltex.additionalRules.word2VecModel.
So in theory, is it possible to train and use my own language model? I really need one in my academic writing!

I'm new to LanguageTool, and found no further information about such model's format here and in dev.languagetool.org.
(The most relevent link may be Finding errors using n-gram data, but the --languagemodel here is just the n-gram data.)

I wonder how the model should be.