1 paper · 1 filter
Indraneil Paul, Haoyi Yang, Goran Glavaš +2
Language models (LMs) have become a staple of the code-writing toolbox. Their pre-training recipe has, however, remained stagnant over recent years, barring the occasional changes…