Chapter 6 - CUDA goes out of memory even with batch size 4. about pytorch-computer-vision-cookbook HOT 4 CLOSED

Is this for chapter 6: Single-Object Segmentation?
I cannot recall the "out of memory message" for Chapter 6 since the model and data are pretty light.
At what stage did you get the error?

from pytorch-computer-vision-cookbook.

Yes 6th chapter. While training as i have 6GB CUDA goes out of memory after 26 epochs, then i used kaggle kernel they have 15GB and CUDA went out of memory after 72 epochs.

from pytorch-computer-vision-cookbook.

My guess is that the error is due these two lines

    pred= torch.sigmoid(output)
    _, metric_b=dice_loss(pred, target)

inside loss_batch helper function. They calculate the dice metric during training.

Can you try one of these two options:

• comment out the two lines and return a fixed number for metric_b for now to see if that is causing the error
• put the two lines under with torch.no_grad(): to avoid tracking gradients when calculating the metric.

from pytorch-computer-vision-cookbook.

Thank you,
i used "with torch.no_grad():" and it works perfectly.
Thank you very much.

from pytorch-computer-vision-cookbook.