[FIX] torch can not use GPU

[andreiramani]

Describe the bug

I have successfully setup khoj using conda, postgresql 16, pytorch=2.1.0, torchvision=0.16.0, torchaudio=2.1.0, pytorch-cuda=12.1.

How to set torch.device to use my GPU?

Screenshots/Error message

UserWarning: Failed to initialize NumPy: DLL load failed while importing _multiarray_umath: The specified module could not be found. (Triggered internally at C:\cb\pytorch_1000000000000\work\torch\csrc\utils\tensor_numpy.cpp:84.)
device: torch.device = torch.device(torch._C._get_default_device()), # torch.device('cpu'),

Platform

• Server:
  • Self-Hosted Python package
• Client:
  • Web browser
• OS:
  • Windows 11, RTX 3050, CUDA runtime 12.1

[andreiramani]

Force upgrade to numpy 2.0.1 gave error start. So, I downgraded to numpy=1.26.4. Relaunch khoj successfully, but it still not detect the GPU.