ConsistentEE: A Consistent and Hardness-Guided Early Exiting Method for Accelerating Language Models Inference

arXiv:2312.11882v2 Announce Type: replace-cross
Abstract: Early Exiting is one of the most popular methods to achieve efficient inference. Current early exiting methods adopt the (weighted) sum of the cross entropy loss of all internal classifiers during training, imposing all these classifiers to predict all instances correctly. However, during inference, as long as one internal classifier predicts an instance correctly, it can accelerate without losing accuracy. Thus, there is a notable gap between training and inference. We propose ConsistentEE, an early …

abstract arxiv classifiers consistent cs.ai cs.cl cs.lg current entropy however inference instances language language models loss popular sum training type