Tag
This paper evaluates the performance of open multimodal large language models on Khmer document VQA, finding that while models handle English and numeric content well, understanding native Khmer script remains challenging. The study uses a diagnostic subset from the KH-FUNSD collection and compares different model configurations.