Open
Description
When i try to train yolov4-crowdhuman dataset with yolov4, loss and avg_loss is always nan, but when i try to train same dataset with yolov4-tiny or yolov4-tiny-3l or even yolov7-tiny its ok.
In addition when i try to train same dataset with same system (Same CUDA and CUDNN version, same laptop) with AlexeyAB repo and yolov4, its working fine.
My System:
NVIDIA GeForce RTX 3070 Mobile
12th Gen Intel i7-12700H
CUDA Version 12.4 with drivers
CUDNN 8.9.7