the.ai

Optimization / Generalization

verified

Compute-Optimal Training

Given a fixed compute budget, you can spend it on a bigger model or on more data. For years the field chose bigger models, and it was wrong: for a given budget the two should grow roughly in step, which means most large models of that era were badly undertrained.

Viz primitive · loss-curvesteps = 1000 · lr = 0.001 · params = 1
loss
step 01000

Loss over 1000 training steps, starting near 6.0. It falls to about 2.16, with 88% of the total improvement arriving in the first half.

Loss over a run at this model size. Drag the parameter count to watch the floor fall — the budget a bigger model burns through faster is not in this curve, and it is what puts the optimum in the middle.

1

Reviewed by opendroid · 2026-08-04