Does avalanche support gradient accumulation?

[man2machine]

I'm running out of memory when training, and would like to use gradient accumulation if possible. Is this supported or is there any easy way to modify the code to enable this?

[AntonioCarta]

the easiest way to do it is to modify the template yourself when the backward is called.