What is Nostr?
Urusan /
npub170e…rnla
2024-08-29 03:11:34

Urusan on Nostr: On a lark, I massively overtrained a model and the results were fascinating. This ...

On a lark, I massively overtrained a model and the results were fascinating.

This wasn't completely random, I heard that there's this "grokking" phenomena when you overtrain a model by something like 10x.

Well, it seems to be real. At the very least it kept learning well past its initial plateau, and it accelerated after a certain point.




Author Public Key
npub170eg5z88cnluedr5sjhz6jm02elv2csaum46qz4g94383ywd478qvcrnla