Generated checkpoint files throwing error when trying to infere an image

[ikram-md]

I recently tried to build my own dataset and train the model. I obtained an error which is attached below.
To investigate the problem further, i tried running a training on 1 single epoch on the original vg dataset. when i tried to preform inference with the generated .pth i obtained the same error that i obtained when i trained the same model on the filtered set :

I am training on 1 GPU, for some reason the values in the .pth file are corrupted.

[ikram-md]

I noticed that i obtain the same error when i tried to run inference on a white image using the pre-trained model provided on the GitHub repository.