JEDI Demo

NLI demo: JEDI decomposes premises into extractive atomic facts and predicts fact-level entailment/contradiction/neutral in one forward pass, without generative LLMs at inference time.

Related paper: Extractive Fact Decomposition for Interpretable Natural Language Inference in One Forward Pass

EMBER Demo

Interactive streaming NER demo: EMBER extracts entity spans from a decoder-only LM's internal representations during generation, enabling simultaneous text generation and tagging with minimal latency overhead.

Related paper: Embedded Named Entity Recognition using Probing Classifiers