You should not finetune the models on the strongest setting of Stockfish as the move will not be understandable unless you really dig deep into the position and

You should not finetune the models on the strongest setting of Stockfish as the move will not be understandable unless you really dig deep into the position and the model would not be able to find a pattern to make sense of it, instead I suggest training on human games of a certain ELO (less than grandmaster).