New Method Enables LLMs to Perform Zero-Shot Named Entity Recognition 20x Faster

Researchers introduced Just Pass Twice (JPT), a technique allowing causal language models to perform token classification with bidirectional context by duplicating input sentences. The method achieves state-of-the-art zero-shot NER results, outperforming previous approaches by 7.9 F1 points while being 20x faster than generative alternatives, eliminating hallucinations and formatting errors.