PaDeLLM-NER: Parallel Decoding in Large Language Models for Named Entity Recognition

TL;DR

This paper introduces PaDeLLM-NER, a parallel decoding method for LLMs that significantly reduces NER inference latency while maintaining high prediction quality, applicable to multiple languages.

Contribution

The paper presents a novel parallel decoding approach for LLM-based NER that integrates seamlessly without extra modules, drastically improving inference speed.

Findings

Inference speed increased by 1.76 to 10.22 times

Maintains state-of-the-art NER performance

Effective for both English and Chinese datasets

Abstract

In this study, we aim to reduce generation latency for Named Entity Recognition (NER) with Large Language Models (LLMs). The main cause of high latency in LLMs is the sequential decoding process, which autoregressively generates all labels and mentions for NER, significantly increase the sequence length. To this end, we introduce Parallel Decoding in LLM for NE} (PaDeLLM-NER), a approach that integrates seamlessly into existing generative model frameworks without necessitating additional modules or architectural modifications. PaDeLLM-NER allows for the simultaneous decoding of all mentions, thereby reducing generation latency. Experiments reveal that PaDeLLM-NER significantly increases inference speed that is 1.76 to 10.22 times faster than the autoregressive approach for both English and Chinese. Simultaneously it maintains the quality of predictions as evidenced by the performance that is on par with the state-of-the-art across various datasets.