2 ms·BabyLM Challenge: Efficient pretraining on a developmentally plausible corpus1 points by whit 3y ago