Maximum submission size is too small in some cases
alexlyzhov opened this issue · comments
Alexander Lyzhov commented
Basically for many applicable neural networks size of weights is larger and may be extremely tough to fit it. Is 2MB limit in place because neural nets are discouraged in the competition or because of some other reason? If it's the latter, I'd invite you to consider pushing it higher.
If neural networks are OK but throwing in a lot of compute in place of clever algorithms is discouraged, 2MB is still low because 200MB weights still seem bound to fit in any single modern GPU or RAM. I'm not that sure about the processor time / GPU time required to train these though.