Recent advancements in large language models (LLMs) have raised concerns about inference costs, increasing the need for research into model compression. While knowledge distillation (KD) is a prominent method for this, research on KD for generative language models like LLMs is relatively sparse, and the approach of distilling student-friendly knowledge, which has shown promising performance in KD for classification models, remains unexplored in generative language models. To explore this approach, we propose PromptKD, a simple yet effective method that utilizes prompt tuning - for the first time in KD - to enable generative language models to transfer student-friendly knowledge. Unlike previous works in classification that require fine-tuning the entire teacher model for extracting student-friendly knowledge, PromptKD achieves similar effects by adding a small number of prompt tokens and tuning only the prompt with student guidance. Extensive experiments on instruction-following datasets show that PromptKD achieves state-of-the-art performance while adding only 0.0007% of the teacher’s parameters as prompts. Further analysis suggests that distilling student-friendly knowledge alleviates exposure bias effectively throughout the entire training process, leading to performance enhancements.

The ability to extract entities and their relations from unstructured text is essential for the automated maintenance of large-scale knowledge graphs. To keep a knowledge graph up-to-date, an extractor needs not only the ability to recall the triples it encountered during training, but also the ability to extract the new triples from the context that it has never seen before. In this paper, we show that although existing extraction models are able to easily memorize and recall already seen triples, they cannot generalize effectively for unseen triples. This alarming observation was previously unknown due to the composition of the test sets of the go-to benchmark datasets, which turns out to contain only 2% unseen data, rendering them incapable to measure the generalization performance. To separately measure the generalization performance from the memorization performance, we emphasize unseen data by rearranging datasets, sifting out training instances, or augmenting test sets. In addition to that, we present a simple yet effective augmentation technique to promote generalization of existing extraction models, and experimentally confirm that the proposed method can significantly increase the generalization performance of existing models.

A computationally expensive and memory intensive neural network lies behind the recent success of language representation learning. Knowledge distillation, a major technique for deploying such a vast language model in resource-scarce environments, transfers the knowledge on individual word representations learned without restrictions. In this paper, inspired by the recent observations that language representations are relatively positioned and have more semantic knowledge as a whole, we present a new knowledge distillation objective for language representation learning that transfers the contextual knowledge via two types of relationships across representations: Word Relation and Layer Transforming Relation. Unlike other recent distillation techniques for the language models, our contextual distillation does not have any restrictions on architectural changes between teacher and student. We validate the effectiveness of our method on challenging benchmarks of language understanding tasks, not only in architectures of various sizes but also in combination with DynaBERT, the recently proposed adaptive size pruning method.