It seems that the the TFT model is running out of memory when preforming hyper parameter tuning need to check if there is a leak somewhere using profiling and comparing how much memory do base models use. Then deciding how to proceed from there.