OpenHalDet: A Unified Benchmark for Hallucination Detection across Diverse Generation Scenarios

OpenHalDet is a unified benchmark for hallucination detection across diverse generation scenarios of large language models. It brings together 17 datasets, 16 detection methods, and 5 backbone LLMs ranging from 3B to 70B parameters, contributed by researchers across Australia, the United States, the United Kingdom, and Singapore. The benchmark is released under the MIT license.