The full-parameter training H20 of Flux Konetxt shows OOM

[RENZ9527]

I trained with the default configuration of Flux Kontext full parameter training (2 H20), and the OOM occurred in less than a minute after the code was executed for training. Maybe I can ask why this is?