Running on DGX Spark

[TuxRobotics]

The DGX Spark requires CUDA 13.0. Here is how I made it run:

Install newer version of torch in the environment:
pip install torch==2.11.0.dev20251215+cu130 torchvision torchaudio triton --index-url https://download.pytorch.org/whl/nightly/cu130

For render.py to run, modify robust.py with:

    input_1 = input
    input_2 = input

    # Convert numpy arrays to tensors if needed
    if isinstance(condition, np.ndarray):
        condition = torch.from_numpy(condition)
    if isinstance(input_1, np.ndarray):
        input_1 = torch.from_numpy(input_1)
    if isinstance(input_2, np.ndarray):
        input_2 = torch.from_numpy(input_2)

And modify gaussians.py with:

    # colors = torch.from_numpy(colors).view(1, -1, 3).float()
    colors = colors if isinstance(colors, torch.Tensor) else torch.from_numpy(colors)
    colors = colors.view(1, -1, 3).float()

Run the render command