Error training on GPU

[Sajjadkhani]

I am trying to train the MS2FP model exactly based on the demo colab and input data from repo. Training works with cpu. Also prediction works with GPU. However when I try training with GPU I get the following error:
Pointer argument (at 2) cannot be accessed from Triton (cpu tensor?)

It runs without error until the following line in this file: IDSL_MINT-main/IDSL_MINT/MS2FP/MS2FP_Model.py:
Source = self.PeakEmbedding(MZ_Tokens, INT).permute(1, 0, 2)

Do you have any idea what can be the problem?
Thanks!

[sajfb]

This is a problem with PyTorch. please see below link:

[issue 1] (#1)

you should downgrade your PyTorch to circumvent this problem or follow up with PyTorch.

[Sajjadkhani]

Seems that the problem is the D_model tensor in line 19 of MS2FP_Model.py.

This solved the problem for me:
`self.device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')

self.D_model = torch.tensor(D_MODEL, device=self.device)`