10 sats \ 0 replies \ @zuspotirko 10 Aug 2023 \ on: Why reproducibility matters | Foundation tech
I remember back in the day there was a famous computer vision neural network that achieved great results at the time but nobody could reproduce it since the randomness seed wasn't saved or deterministic. They published the weights but nodody could achieve the same weights or a model close to that performance
Can't remember what it was called - something similar to imgNet or photoNet or something
But yeah, for pure builds without ML there are plenty of CI/CD solutions helping out with this