I was training yolov5 with custom dataset and got the error in the title. The command I used:

$ python train.py --img 640 --batch 16 --epochs 3 --data D:\custom\data.yaml --weights yolov5s.pt 

I have python 3.9, pytorch 1.11 and cuda 11.7

CodePudding user response：

Please check your cuDNN version. CUDA 11.7 seems like a very recent update and cuDNN may have problem with it. Try use CUDA 11.6 instead, with cuDNN 8.4.1.