Not recognizing libcudart when running commands

I have a GPU compute (Standard_NC6s_v3), and I want to train a model using that. The kernel/environment (Azure ML 3.8) has tensorflow 2.3 install by default, and CUDA 11. But when I run a few commands (example: python3 -c "import tensorflow as tf;print(tf.__version__)" to get the TF version), it gives me a warning:

```
2021-11-16 13:23:10.274930: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcudart.so.10.1'; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory
2021-11-16 13:23:10.274970: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.

```

Why is it looking for libcudart.so.10.1 when version 11 is installed? I also can't manually install CUDA 10.1. I tried resetting the LD_LIBRARY_PATH variable, using `export LD_LIBRARY_PATH=/usr/local/cuda-11.1/targets/x86_64-linux/lib` but that didn't work either.

Checking `ls /usr/local/ `using the terminal, I have three cuda folders; cuda, cuda 11, and cuda 11.1.

How can I bypass that error/warning?

Commencez par la commande signalée `python3 -c "import tensorflow as tf; print(tf.__version__)"` et comparez le runtime CUDA attendu par TensorFlow 2.3 avec l’environnement Azure ML 3.8. Vérifiez les chemins `/usr/local/cuda`, `/usr/local/cuda-11` et `/usr/local/cuda-11.1`, ainsi que `LD_LIBRARY_PATH` ; le travail est considéré comme terminé lorsque l’écart de versions est expliqué et qu’une méthode vérifiée pour supprimer ou éviter l’avertissement est documentée.

Rédigé par le modèle d'indexation à partir du texte de l'issue.