Just Pass Twice: Efficient Token Classification with LLMs for Zero-Shot NER

Name: Just Pass Twice: Efficient Token Classification with LLMs for Zero-Shot NER
Uploaded: 2026-08-26T13:35:04.965959+00:00
Duration: 351 s
Description: The episode discusses 'Just Pass Twice,' a paper presenting an efficient method for Named Entity Recognition (NER) using Large Language Models. The technique solves limitations in LLM causal attention by duplicating input sequences, allowing future context to flow back into the first pass. This results in significant F1 score increases and 20x faster performance than generative methods.

The episode discusses 'Just Pass Twice,' a paper presenting an efficient method for Named Entity Recognition (NER) using Large Language Models. The technique solves limitations in LLM causal attention by duplicating input sequences, allowing future context to flow back into the first pass. This results in significant F1 score increases and 20x faster performance than generative methods.

Open the full episode