when training finishes it hits the memory error

[jakubczakon]

Running train crashes when the pipelines are collating results from running transform on the entire dataset.

[animeshsahu80]

did you find any solution

[jakubczakon]

Well, the training is working and the checkpoints are saved after every epoch (if improved) so the trained model is available.

All that needs to be done is:

cp experiment/checkpoints/unet/best.torch experiment/transformers/unet

and then run the evaluation/prediction.

So it is more of an inconvenience than a big problem.