Torch 1.13.0 + CUDA 11.6 error when import _C

[hungdche]

Hi,

I'm running into this error while trying to install nksr. My torch version is 1.13.0, and my CUDA version is 11.6, so I download nksr using the following command:

pip install -U nksr -f https://nksr.huangjh.tech/whl/torch-1.13.0+cu116.html

However, when I ran python examples/recons_waymo.py, I ran into this error

Traceback (most recent call last):
  File "/.../NKSR/examples/recons_waymo.py", line 10, in <module>
    import nksr
  File "/.../.local/lib/python3.9/site-packages/nksr/__init__.py", line 18, in <module>
    from nksr.nn.unet import SparseStructureNet
  File "/.../.local/lib/python3.9/site-packages/nksr/nn/__init__.py", line 10, in <module>
    from .modules import Conv3d, GroupNorm, Activation, GroupNorm, MaxPooling, Upsampling, SparseZeroPadding
  File "/.../.local/lib/python3.9/site-packages/nksr/nn/modules.py", line 14, in <module>
    from nksr.svh import SparseFeatureHierarchy, KernelMap, VoxelStatus
  File "/.../.local/lib/python3.9/site-packages/nksr/svh.py", line 14, in <module>
    from nksr.ext import _CpuIndexGrid, _CudaIndexGrid
  File "/.../.local/lib/python3.9/site-packages/nksr/ext.py", line 9, in <module>
    from nksr import _C
ImportError: /.../.local/lib/python3.9/site-packages/nksr/_C.so: undefined symbol: _ZN2at4_ops5zeros4callEN3c108ArrayRefINS2_6SymIntEEENS2_8optionalINS2_10ScalarTypeEEENS6_INS2_6LayoutEEENS6_INS2_6DeviceEEENS6_IbEE

I check my torch version with python -c "import torch; print(torch.__version__)"

1.13.0.dev20220616

and cuda version with nvcc --version

Copyright (c) 2005-2022 NVIDIA Corporation
Built on Thu_Feb_10_18:23:41_PST_2022
Cuda compilation tools, release 11.6, V11.6.112
Build cuda_11.6.r11.6/compiler.30978841_0

Any help is much appreciated. Thanks!

[heiwang1997]

Hey @hungdche your download command should be correct. The 1.13.0.dev20220616 is a bit suspicious though. Did you install torch in a standard way (such as pip or conda)?

[hungdche]

Hi. I'm not sure since torch was preinstalled. I updated to torch2.0.0 and it works now. Thanks!