RuntimeError: Trying to create tensor with negative dimension -920509440: [-920509440]

[hammockrobotics]

i keep getting the error at the 3rd epoch. it says cannot create tensor with negative dimension:
I used all the settings untouched as per the repo mentioned. please assist as i have tried the following already:

• Editing the dataset to a completely new one from Roboflow thinking my dataset was corrupted.
• Renaming the dataset labels and images to simpler 01.jpg - 01.txt , 02.jpg - 02.txt and so on...

so I believe my dataset should not be the issue as i have downloaded it and tested it with multiple sources from roboflow. I used the format yolo-scaled.

I am also using the mish wewights for implementation. Where can I start to look regarding finding the root cause of the issue? Appreciate the help greatly. error as below from the 3rd epoch before failing:

  3/99     9.67G   0.09145     2.622   0.01558     2.729         3       640: 100%|█| 250/250 [01:27<00:00,  3.18it/s]
           Class      Images     Targets           P           R      [email protected]  [email protected]:.95:   0%|                                                                                 | 0/33 [00:00<?, ?it/s]

Traceback (most recent call last):
File "train.py", line 537, in
train(hyp, opt, device, tb_writer, wandb)
File "train.py", line 344, in train
log_imgs=opt.log_imgs if wandb else 0)
File "/yolo/test.py", line 134, in test
output = non_max_suppression(inf_out, conf_thres=conf_thres, iou_thres=iou_thres)
File "/yolo/utils/general.py", line 340, in non_max_suppression
i = torch.ops.torchvision.nms(boxes, scores, iou_thres)
RuntimeError: Trying to create tensor with negative dimension -920509440: [-920509440]