Markdown supported · ⌘↵ to post
this dope, had me laughing reading this, but one thing though, the reason why one epoch is enough it's because of computing power, we u have large data sets it's only reasonable to train your llm on half epoch or 1.2 epoch, to save computing power, and it's the standard of training llms
Dominic Oladapo-Tonade
John Raphael
This is funny and at the same time knowledgeable, it’s nice to always try things out when building models , you never can tell what the best epoch or parameter is