XtremeDistilTransformers: Task Transfer for Task-agnostic Distillation
This work addresses the challenge of deploying large NLP models in resource-limited environments by providing a more efficient compression method, though it is incremental as it builds on existing distillation techniques.
The authors tackled the problem of compressing large pre-trained language models for resource-constrained settings by developing XtremeDistilTransformers, a task-agnostic distillation framework that leverages task-specific methods to create small universal models, achieving state-of-the-art performance on tasks like GLUE, SQuAD, and multi-lingual NER with models of 13MM, 22MM, and 33MM parameters.
While deep and large pre-trained models are the state-of-the-art for various natural language processing tasks, their huge size poses significant challenges for practical uses in resource constrained settings. Recent works in knowledge distillation propose task-agnostic as well as task-specific methods to compress these models, with task-specific ones often yielding higher compression rate. In this work, we develop a new task-agnostic distillation framework XtremeDistilTransformers that leverages the advantage of task-specific methods for learning a small universal model that can be applied to arbitrary tasks and languages. To this end, we study the transferability of several source tasks, augmentation resources and model architecture for distillation. We evaluate our model performance on multiple tasks, including the General Language Understanding Evaluation (GLUE) benchmark, SQuAD question answering dataset and a massive multi-lingual NER dataset with 41 languages. We release three distilled task-agnostic checkpoints with 13MM, 22MM and 33MM parameters obtaining SOTA performance in several tasks.