Multi-stage Distillation Framework for Cross-Lingual Semantic Similarity Matching
This work addresses the deployment challenge for cross-lingual semantic similarity matching in resource-constrained environments, representing an incremental improvement over existing distillation methods.
The paper tackles the problem of deploying large cross-lingual models on memory-limited devices by proposing a multi-stage distillation framework that compresses models like XLM-R and MiniLM by over 50% while reducing performance by only about 1%.
Previous studies have proved that cross-lingual knowledge distillation can significantly improve the performance of pre-trained models for cross-lingual similarity matching tasks. However, the student model needs to be large in this operation. Otherwise, its performance will drop sharply, thus making it impractical to be deployed to memory-limited devices. To address this issue, we delve into cross-lingual knowledge distillation and propose a multi-stage distillation framework for constructing a small-size but high-performance cross-lingual model. In our framework, contrastive learning, bottleneck, and parameter recurrent strategies are combined to prevent performance from being compromised during the compression process. The experimental results demonstrate that our method can compress the size of XLM-R and MiniLM by more than 50\%, while the performance is only reduced by about 1%.