Asked about Predicting capability from compute
Why do deployed models often exceed compute-optimal training?
An answer that works
Compute-optimal minimises training cost. If inference volume is large, inference dominates total cost, so a smaller model trained well past that point is cheaper overall despite costing more to train.
Reading an answer is not being able to give one. The section that teaches this is Predicting capability from compute, 7 hours of Scaling laws.
Learn this properly