Trying to be a little less negative than the guy that got flagged, I too don't understand the motivation behind these projects. I've seen a hundred of them at this point- and each of them is probably worse and has less learning value than the one Andrej Karpathy made to teach people the building blocks involved in a GPT
So, why do people keep making these? Asking genuinely
Trying to be a little less negative than the guy that got flagged, I too don't understand the motivation behind these projects. I've seen a hundred of them at this point- and each of them is probably worse and has less learning value than the one Andrej Karpathy made to teach people the building blocks involved in a GPT
So, why do people keep making these? Asking genuinely
"hn1g.txt" => not available
Interesting trend of labeling models after how big they are on disk vs how many learnable parameters they have.
At that scale it seems like you could just solve the KKT conditions directly. Exaggerating but only a little