halueval

A hallucination detection pipeline for Large Language Models (LLMs).

Framework for evaluating LLM hallucination detection in Department of Defense knowledge domains. Implements HaluEval, multi-method generation, and ensemble detection.

Privacy-aware enterprise knowledge intelligence. Ask a question over your organisation's documents and get an answer in which every sentence has been checked against a source you are allowed to read — or no answer at all

Lightweight, CPU-only benchmark of five hallucination-detection methods (ROUGE-L, semantic similarity, BERTScore, NLI, and an ensemble) across the three HaluEval tasks: QA, dialogue, and summarisation. Code for arXiv:2606.29809.

Adaptive Activation Cancellation (AAC): real-time hallucination suppression in LLMs via H-Node identification and confidence-weighted forward hooks. No fine-tuning, no external knowledge, zero capability degradation. Tested on OPT-125M, Phi-3-mini, LLaMA 3-8B.