Could not load dynamic library 'libcusolver.so.10'; dlerror

[DavidGOrtega]

Coming from here

Could not load dynamic library 'libcusolver.so.10'; dlerror: libcusolver.so.10: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/lib/python3.8/dist-packages/cv2/../../lib64:/usr/local/nvidia/lib:/usr/local/nvidia/lib64

[0x2b3bfa0]

It's not a bug; it's a feature. 😉 Our images use CUDA 11, not CUDA 10.

$ docker run --rm iterativeai/cml:0-dvc2-base1-gpu find / -name libcusolver.so.*
/usr/local/cuda-11.2/targets/x86_64-linux/lib/libcusolver.so.11
/usr/local/cuda-11.2/targets/x86_64-linux/lib/libcusolver.so.11.1.0.152

[0x2b3bfa0]

If you need CUDA 10, you can use the official nvidia/cuda:10.1-cudnn7-runtime-ubuntu18.04 base image and install our tools on top of it:

CML

apt update && apt install -y curl && curl https://github.com/iterative/cml/releases/latest/download/cml-linux-x64 -Lo /usr/bin/cml && chmod a+x /usr/bin/cml

DVC

apt update && apt install -y curl && curl https://dvc.org/deb/dvc.list -Lo /etc/apt/sources.list.d/dvc.list && curl -L https://dvc.org/deb/iterative.asc | apt-key add - && apt update && apt install -y dvc

[0x2b3bfa0]

We've seen a few requests in the past regarding CUDA versions:

• upgrade cuda version to support tf 2.6.0 #766
• Docker with latest version of Cuda #685

Should we take another look at packaging?

[mevatron]

Upgrading to tensorflow~=2.6.0 does the trick! Thanks again!