replace nnunetv2's with MedNeXt v1 have some problems

[yml-bit]

Hi, I replaced the model in nnunetv2 with MedNeXt v1 and trained the conveyor farmer i function as a nan value, and I see that the original nnunetv2 raw output is 3 convolution, and after adding a 3d convolution to the deep supervised output of MedNeXt v1, the situation is not improved!