Fetching the paper…

PYRA: Parallel Yielding Re-Activation for Training-Inference Efficient Task Adaptation · Around