Knowledge Distillation of Russian Language Models with Reduction of Vocabulary
This work addresses computational efficiency for industrial NLP applications in Russian, offering a novel compression method that is incremental but provides significant practical gains.
The paper tackled the problem of vocabulary mismatch in knowledge distillation for Russian language models, proposing alignment techniques that achieved compression ratios from 17x to 49x while maintaining quality comparable to a 1.7x compressed model with full vocabulary.
Today, transformer language models serve as a core component for majority of natural language processing tasks. Industrial application of such models requires minimization of computation time and memory footprint. Knowledge distillation is one of approaches to address this goal. Existing methods in this field are mainly focused on reducing the number of layers or dimension of embeddings/hidden representations. Alternative option is to reduce the number of tokens in vocabulary and therefore the embeddings matrix of the student model. The main problem with vocabulary minimization is mismatch between input sequences and output class distributions of a teacher and a student models. As a result, it is impossible to directly apply KL-based knowledge distillation. We propose two simple yet effective alignment techniques to make knowledge distillation to the students with reduced vocabulary. Evaluation of distilled models on a number of common benchmarks for Russian such as Russian SuperGLUE, SberQuAD, RuSentiment, ParaPhaser, Collection-3 demonstrated that our techniques allow to achieve compression from $17\times$ to $49\times$, while maintaining quality of $1.7\times$ compressed student with the full-sized vocabulary, but reduced number of Transformer layers only. We make our code and distilled models available.