Multistep weights in forecast loss not applied

[robert-DL]

Hi Folks,

The issue is about the code snippet here:
https://github.com/INRIA/geoarches/blob/929d6b2364fa29dc14fdd0c6ac80cae37ef293e5/geoarches/lightning_modules/forecast.py#L166C13-L166C59

td.apply() is not in-place!(https://docs.pytorch.org/tensordict/stable/reference/generated/tensordict.TensorDict.html#tensordict.TensorDict.apply)
The desired in-place operation is td.apply_().
I therefore expect that the multistep weights are not applied at all (but I did not test for it). I just checked the tensordict documentation and tried a dummy example. In the example my hypothesis holds true.

Best
Robert