conftrace_

2022 EMNLP EMNLP 2022

Fast Vocabulary Transfer for Language Model Compression

Abstract

AbstractReal-world business applications require a trade-off between language model performance and size. We propose a new method for model compression that relies on vocabulary transfer. We evaluate the method on various vertical domains and downstream tasks. Our results indicate that vocabulary transfer can be effectively used in combination with other compression techniques, yielding a significant reduction in model size and inference time while marginally compromising on performance.

🐣 Hot Topic Early Bird - inference optimization

Authors

Leonidas Gee, Andrea Zugarini, Leonardo Rigutini, Paolo Torroni

Topics

Artificial Intelligence > Core AI > Model Compression Machine Learning > Application Areas > Knowledge Distillation Deep Learning > Optimization & Theory > Model Compression Deep Learning > Learning Types > Transfer Learning Deep Learning > Models > Language Models

Keywords

model compression parameter efficient language model parameter-efficient fine-tuning inference optimization inference time vocabulary transfer

More from EMNLP 2022

All EMNLP 2022 papers →

Improving Precancerous Case Characterization via Transformer-based Ensemble Learning

Developing Prefix-Tuning Models for Hierarchical Text Classification

PAIGE: Personalized Adaptive Interactions Graph Encoder for Query Rewriting in Dialogue Systems

Multimodal Context Carryover

Distilling Multilingual Transformers into CNNs for Scalable Intent Classification

Bringing the State-of-the-Art to Customers: A Neural Agent Assistant Framework for Customer Service Support