BERT Loses Patience: Fast and Robust Inference with Early Exit
This addresses efficiency and robustness issues for users of large language models, though it is incremental as it builds on existing early exit techniques.
The paper tackles the problem of inefficient and non-robust inference in pretrained language models by introducing Patience-based Early Exit, a plug-and-play method that dynamically stops inference when intermediate predictions stabilize, resulting in improved accuracy and faster inference with fewer layers.
In this paper, we propose Patience-based Early Exit, a straightforward yet effective inference method that can be used as a plug-and-play technique to simultaneously improve the efficiency and robustness of a pretrained language model (PLM). To achieve this, our approach couples an internal-classifier with each layer of a PLM and dynamically stops inference when the intermediate predictions of the internal classifiers remain unchanged for a pre-defined number of steps. Our approach improves inference efficiency as it allows the model to make a prediction with fewer layers. Meanwhile, experimental results with an ALBERT model show that our method can improve the accuracy and robustness of the model by preventing it from overthinking and exploiting multiple classifiers for prediction, yielding a better accuracy-speed trade-off compared to existing early exit methods.