Experiment: Effect of Model Size

tldr; Huge gains upto small, marginal after base

Outcome

In this experiment, I trained and evaluated four different model sizes:

1. xsmall (green): CV 0.686, LB 0.738
2. small (pink): CV 0.706, LB 0.752
3. base (blue): CV 0.7391, LB 0.755
4. large (yellow): CV ???, LB ???