An Empirical Study of Leveraging Knowledge Distillation for Compressing Multilingual Neural Machine Translation Models
This addresses the problem of model size for multilingual translation users, but it is incremental as it builds on existing knowledge distillation methods.
The paper tackled compressing multilingual neural machine translation models using knowledge distillation, finding that common approaches yield 4-5x smaller models but suffer performance drops of up to 3.5 BLEU, with deeper compact models performing similarly to shallower non-compact ones.
Knowledge distillation (KD) is a well-known method for compressing neural models. However, works focusing on distilling knowledge from large multilingual neural machine translation (MNMT) models into smaller ones are practically nonexistent, despite the popularity and superiority of MNMT. This paper bridges this gap by presenting an empirical investigation of knowledge distillation for compressing MNMT models. We take Indic to English translation as a case study and demonstrate that commonly used language-agnostic and language-aware KD approaches yield models that are 4-5x smaller but also suffer from performance drops of up to 3.5 BLEU. To mitigate this, we then experiment with design considerations such as shallower versus deeper models, heavy parameter sharing, multi-stage training, and adapters. We observe that deeper compact models tend to be as good as shallower non-compact ones, and that fine-tuning a distilled model on a High-Quality subset slightly boosts translation quality. Overall, we conclude that compressing MNMT models via KD is challenging, indicating immense scope for further research.