More relevantly, it means that the model parameter count, dataset size, and training procedure are balanced such that they can achieve equal or better performan

More relevantly, it means that the model parameter count, dataset size, and training procedure are balanced such that they can achieve equal or better performance as GPT-3 while being smaller.