Finding Experts in Transformer Models (2005.07647v1)

Abstract: In this work we study the presence of expert units in pre-trained Transformer Models (TM), and how they impact a model's performance. We define expert units to be neurons that are able to classify a concept with a given average precision, where a concept is represented by a binary set of sentences containing the concept (or not). Leveraging the OneSec dataset (Scarlini et al., 2019), we compile a dataset of 1641 concepts that allows diverse expert units in TM to be discovered. We show that expert units are important in several ways: (1) The presence of expert units is correlated ($r^2=0.833$) with the generalization power of TM, which allows ranking TM without requiring fine-tuning on suites of downstream tasks. We further propose an empirical method to decide how accurate such experts should be to evaluate generalization. (2) The overlap of top experts between concepts provides a sensible way to quantify concept co-learning, which can be used for explainability of unknown concepts. (3) We show how to self-condition off-the-shelf pre-trained LLMs to generate text with a given concept by forcing the top experts to be active, without requiring re-training the model or using additional parameters.