gpu memory issue

[babbu3682]

I have a problem with the Encoder model. I wanted to make the CRNN from scratch so I build my custom CNN encoder model like "Efficientnet-b0", it is worked but after 2 steps the gpu memory errors show up. I think GPU memory is stacked every step because of "for() function" in forward(). So, how can I solve that problem?

1 epoch

2 epoch

3 epoch

4 epoch

as you can see above, the memory increase until CUDA out of error .....

[HHTseng]

Sorry that I was not able to conceive where the problem may be without knowing more details how your new model "Efficientnet-b0" is constructed. I do not recognize this type of error from my experience.