Support for PyTorch 2.6+ to enable NVIDIA 50-series GPU compatibility

[luyida1999]

Hi there,

I would like to request an update to support PyTorch 2.6+.

With the upcoming release of NVIDIA 50-series GPUs (e.g., RTX 5090), users will likely need to upgrade to CUDA 12.8+ and subsequently PyTorch 2.6+ to utilize the new hardware.

Thank you for your hard work on this project!

[HarjoteNest]

Solution

After creating the Conda environment, install the package in editable mode:

pip install -e .

Then run the following commands to fix script formatting and permissions:

sed -i 's/\r$//' bin/C2C
chmod +x bin/C2C

Verify that the environment is set up correctly by running:

bin/C2C --help

If the help message is displayed, the environment is configured successfully.

For systems using NVIDIA RTX 50-series GPUs, update the PyTorch version as follows:

pip uninstall -y torch torchvision torchaudio
pip install torch==2.7.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

If you encounter issues related to loading weights from an older PyTorch version, run the following (for the current terminal session):

export TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1

After this, running any pipeline should work correctly, though you may see some non-blocking warnings. For example:

bin/C2C spine_muscle_adipose_tissue -i "dicom" -o "output"

---

Thank you.