1 paper
Andrei Piterbarg, Sean McLeish, Dhruv Pai +1
Looped transformers reuse the same layer multiple times, allowing inference compute to be adjusted without changing the model. Recent adaptive-depth methods train the model across…