An error was reported for training the model,
nvidia-smi checked the graphics card driver and found that CUDA version: ERR!
I searched for related solutions, and most of them said that it was a mismatch between CUDA, cuDNN, and graphics card driver versions, but I have used this version for a long time, and this problem has never occurred, and then I checked the problem and suspected it was CUDA's dynamic link library problem, checked the dynamic link library files in the CUDA/lib directory, and there were two more files,
After deleting these two files, nvidia-smi displays normally