Ilya Emelyanov@KantervileGhostOct 4Aleph Alpha just dropped Kolibri: 78B params, 3B active. Matches models 4x its size. Which raises a question everyone's avoiding: Why are we still racing for bigger models? 🧵17
Ilya Emelyanov@KantervileGhostOct 4I build ML models for small data. When n < 1000, standard tools get weaker. So I wrote my own. SmallGBM and SmallMLP Both open source. Check it here: github.com/nsdmlk/SmallGBM github.com/nsdmlk/SmallMLPGitHub - nsdmlk/SmallGBM: SmallGBM is a gradient boosting library designed for small datasets (n <...From github.com9
