Import Error when installing mmcv using pip and mim

Prerequisite

[X] I have searched Issues and Discussions but cannot get the expected help.
[X] The bug has not been fixed in the latest version(https://github.com/open-mmlab/mmcv).

Environment

I am trying to install mmaction2 in my Google Colab environment by following the official google Colab tutorial (https://colab.research.google.com/github/open-mmlab/mmaction2/blob/master/demo/mmaction2_tutorial.ipynb#scrollTo=No_zZAFpWC-a)

This is my Google Colab environment:

nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2023 NVIDIA Corporation Built on Tue_Aug_15_22:02:13_PDT_2023 Cuda compilation tools, release 12.2, V12.2.140 Build cuda_12.2.r12.2/compiler.33191640_0 gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 Copyright (C) 2021 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

In the tutorial, PyTorch and torch vision use CUDA 11.8:

%pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

and then installing MMEngine:

%pip install -U openmim
!mim install mmengine

and mmcv is installed like this:

!mim install "mmcv==2.0.0" However, then I get compatibility issues with the CUDA version of the Google Colab (12.1). Then, I tried to install the latest PyTorch version for CUDA 12.1 and the newest mmcv version (as described on the website):

!pip install mmcv==2.1.0 -f https://download.openmmlab.com/mmcv/dist/cu121/torch2.1/index.html mmaction2 is installed like this:

# Install mmaction2
!rm -rf mmaction2
!git clone https://github.com/open-mmlab/mmaction2.git -b main
%cd mmaction2

!pip install -e .

# Install some optional requirements
!pip install -r requirements/optional.txt

Reproduces the problem - code sample

https://colab.research.google.com/github/open-mmlab/mmaction2/blob/master/demo/mmaction2_tutorial.ipynb#scrollTo=No_zZAFpWC-a

import torch, torchvision
print(torch.__version__, torch.cuda.is_available())

import mmaction
print(mmaction.__version__)

from mmcv.ops import get_compiling_cuda_version, get_compiler_version
print(get_compiling_cuda_version())
print(get_compiler_version())

from mmengine.utils.dl_utils import collect_env
print(collect_env())

Reproduces the problem - command or script

# Check Pytorch installation
import torch, torchvision
print(torch.__version__, torch.cuda.is_available())

# Check MMAction2 installation
import mmaction
print(mmaction.__version__)

# Check MMCV installation
from mmcv.ops import get_compiling_cuda_version, get_compiler_version
print(get_compiling_cuda_version())
print(get_compiler_version())

# Check MMEngine installation
from mmengine.utils.dl_utils import collect_env
print(collect_env())

Reproduces the problem - error message

2.2.1+cu121 True
1.2.0
---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
[<ipython-input-2-55109e865b0d>](https://bhzb0mtqv64-496ff2e9c6d22116-0-colab.googleusercontent.com/outputframe.html?vrz=colab_20240411-060128_RC00_623759654#) in <cell line: 10>()
      8 
      9 # Check MMCV installation
---> 10 from mmcv.ops import get_compiling_cuda_version, get_compiler_version
     11 print(get_compiling_cuda_version())
     12 print(get_compiler_version())

3 frames
[/usr/lib/python3.10/importlib/__init__.py](https://bhzb0mtqv64-496ff2e9c6d22116-0-colab.googleusercontent.com/outputframe.html?vrz=colab_20240411-060128_RC00_623759654#) in import_module(name, package)
    124                 break
    125             level += 1
--> 126     return _bootstrap._gcd_import(name[level:], package, level)
    127 
    128 

ImportError: /usr/local/lib/python3.10/dist-packages/mmcv/_ext.cpython-310-x86_64-linux-gnu.so: undefined symbol: _ZN2at4_ops10zeros_like4callERKNS_6TensorEN3c108optionalINS5_10ScalarTypeEEENS6_INS5_6LayoutEEENS6_INS5_6DeviceEEENS6_IbEENS6_INS5_12MemoryFormatEEE

Additional information

I have tried different versions of mmcv and mmcv-full, different PyTorch versions, but nothing seems to work, e.g.:

from the official mmcv documentation (https://mmcv.readthedocs.io/en/v1.7.0/get_started/installation.html):

!pip install mmcv==2.1.0 -f https://download.openmmlab.com/mmcv/dist/cu121/torch2.1/index.html

Nightly CUDA builds (from https://github.com/pytorch/pytorch/issues/91122):

pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu121

open-mmlab / mmcv