CrossCult-KIBench is a multimodal benchmark for evaluating cultural knowledge insertion in multimodal large language models. The benchmark tests whether a model can absorb culture-specific updates while preserving generalization and locality across visually grounded questions.
This data package is intended to be used with the companion anonymized code package at
https://github.com/crosscult-kibench/CrossCult-KIBench. The code package contains environment setup… See the full description on the dataset page:
https://huggingface.co/datasets/crosscult-kibench/CrossCult-KIBench.
