MPS float64 fix merged for 02_predict (HT_H23B) but 01_train (HT_H23B + HT_LIF24) still fails with float64 buffers on MPS

[mm2971-art]

Hi MicroSplit team,
First of all, thank you — the recently merged fix“mps float64 h23b fix (#23)” resolves the MPS float64 error for the HT_H23B 02_predict notebook on my Mac. However, I still encounter the same underlying issue when running training notebooks (01_train) for HT_H23B and HT_LIF24: training crashes when moving the model to the MPS device due to float64 tensors/buffers.
Best,
Markus