Out of memory error when finetuning is enabled about neuraltalk2 HOT 4 CLOSED

4GB is quite little unfortunately. See if it helps to decrease the batch_size ? Try down to 1 to see what happens.

from neuraltalk2.

batch_size 4 seems to do the trick, thank you so much!

from neuraltalk2.

Great! You might want to slightly decrease the learning rate, as a good heuristic, when decreasing the batch size. E.g. reduction by x4 => learning rate reduce ~2-4x is decent guess. Closing, thanks for creating the issue!

from neuraltalk2.

Thanks for the hint and for all your help.
Take care!

Στις Τρίτη, 1 Δεκεμβρίου 2015, ο χρήστης Andrej [email protected]
έγραψε:

Great! You might want to slightly decrease the learning rate, as a good
heuristic, when decreasing the batch size. E.g. reduction by x4 => learning
rate reduce ~2-4x is decent guess. Closing, thanks for creating the issue!

—
Reply to this email directly or view it on GitHub
#35 (comment)
.

from neuraltalk2.