Compute-efficient pretraining and scaling to trillion-parameter models

by ronfriedhaber | View on Hacker News