[Bug]ModuleNotFoundError: No module named 'mmcv._ext'(Buid from source on Linux with CUDA12.8, python 3.12, pytorch2.9.1+cu128 )

[yshawen]

Prerequisite

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

Environment

Linux, CUDA12.8, python 3.12.12, torch 2.9.1+cu128

Reproduces the problem - code sample

python .dev_scripts/check_installation.py --> raise error "No module named 'mmcv._ext'"

Reproduces the problem - command or script

I was building mmcv from source following the next steps according to documents:

1. git clone https://github.com/open-mmlab/mmcv.git
2. cd mmcv
3. pip install -r requirements/optional.txt
4. nvcc --version --> V12.6.20
5. gcc --version --> 12.1.0
6. pip install -e . -v --> Successfully installed mmcv-2.2.0 mmengine-0.10.7
7. python .dev_scripts/check_installation.py --> raise error "No module named 'mmcv._ext'"

Reproduces the problem - error message

Detailed error message after running "python .dev_scripts/check_installation.py":
Traceback (most recent call last):
File "/home/mmcv/.dev_scripts/check_installation.py", line 4, in
from mmcv.ops import box_iou_rotated
File "/home/mmcv/mmcv/ops/init.py", line 3, in
from .active_rotated_filter import active_rotated_filter
File "/home/mmcv/mmcv/ops/active_rotated_filter.py", line 10, in
ext_module = ext_loader.load_ext(
^^^^^^^^^^^^^^^^^^^^
File "/home/mmcv/mmcv/utils/ext_loader.py", line 13, in load_ext
ext = importlib.import_module('mmcv.' + name)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/envs/sam3/lib/python3.12/importlib/init.py", line 90, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ModuleNotFoundError: No module named 'mmcv._ext'

Additional information

No response

[YoushaaMurhij]

I have the same problem building from source on Linux with CUDA11.8, python 3.10, pytorch2.0+cu118

[perfyperfect]

I have the same problem building from source on Linux with CUDA12.1, python 3.10, pytorch2.3.0+cu121

[dina-khalid]

same for CUDA 11.8 PyTorch 2.4

I got the main error when running pip install mmcv==2.2.0 -f https://download.openmmlab.com/mmcv/dist/cu118/torch2.4/index.html
the gpu version isn't installed and i keep getting
WARNING: Retrying (Retry(total=4, connect=None, read=None, redirect=None, status=None)) after connection broken by 'ConnectTimeoutError(<pip._vendor.urllib3.connection.HTTPSConnection object at 0x7fbda90a9190>, 'Connection to download.openmmlab.com timed out. (connect timeout=15)')': /mmcv/dist/cu118/torch2.4/index.html

[Wjqboss]

I have the same problem building from source on Linux with CUDA12.8, python 3.10.19, pytorch2.9.1+cu128

[daker11123]

Linux, CUDA12.8, python 3.12.12, torch 2.9.1+cu128,In the same environment, first upgrade setuptools and wheel: pip install --upgrade setuptools wheel. Then, compile and install using the legacy method: python3 setup.py install. That should work.