Eric J. Michaud comments on QAPR 5: grokking is maybe not *that* big a deal?

[Eric J. Michaud]

Huh those batch size and learning rate experiments are pretty interesting!