Import-time error: undefined symbol: cudaPeekAtLastError

[jantman]

I'm on Debian Buster (10.6) with Python 3.7.3 and nvidia-cuda-dev 9.2.148-7 installed from debian non-free.

I've finally been able to compile and install with GPU support (GPU=1 pip3 install darknetpy) after digging through a number of other issues on here, mainly removing the Debian Rust packages, installing rustup via the shell script installer, and then rustup default nightly.

The package compiles and installs via pip without error. However:

# python3
Python 3.7.3 (default, Jul 25 2020, 13:03:44) 
[GCC 8.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from darknetpy.detector import Detector
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python3.7/dist-packages/darknetpy/__init__.py", line 1, in <module>
    from .darknetpy import *
ImportError: /usr/local/lib/python3.7/dist-packages/darknetpy/darknetpy.cpython-37m-x86_64-linux-gnu.so: undefined symbol: cudaPeekAtLastError
>>>

I'm unable to find any solution to this. Current darknet master itself compiles fine on my system, and runs fine using the ./darknet demo script.

Any advice would be greatly appreciated...

[sklett-src]

+1 Same Problem, were you able to solve it? somehow a normal shell finds cuda at the right place, but python doesnt.