got Out of GPU memory when learning

[A-Cepheus]

It seems like one iteration can be completed, but OOM occurred during the second iteration, any idea?

[A-Cepheus]

Maybe I should continue reduce batch size?

[jonathan-laurent]

You should probably reduce batch size indeed.

[A-Cepheus]

now got a new error

[jonathan-laurent]

Out of memory errors are often shown as other errors. I would reduce the batch size and/or network size even further.

[A-Cepheus]

I feel that the problem with OOM is indeed accompanied by mem_ buff Appearing as the size increases.

[A-Cepheus]

This is a possible reason that I am researching: FluxML/FluxTraining.jl#148