Getting CUDA memory error!

Continuing the discussion from Lesson 3 In-Class Discussion :white_check_mark::

I am trying the planet data - it works fine in 128. However when running lr_find with 256 I run into out of memory. How to solve this?


1 Like

You’re running out of memory, try halving your batch size

1 Like

before you resize the batch size, close your kernel, and then run it again

Thanks. Seems to work - set the batch size to 32 ( I guess the default was 64).

However takes a long time to train - around 7min / iteration.

@polysoulz - thanks I continued running without kernel restart. I seems to work for me.