Update README.md

README.md

@@ -57,7 +57,7 @@ This README provides an overview of the dataset and instructions on how to use i
 
 ## Dataset Overview
 
-DocAttriBench is a dataset for training and evaluating answer grounding in Document Visual Question Answering. The dataset is constructed from existing document VQA and document understanding datasets. From these source datasets, we use Docling to extract document layout elements and their semantic types, and **MAPPET** to obtain the bounding box of the evidence region supporting the answer.
+DocAttriBench is a dataset for training and evaluating answer grounding in Document Visual Question Answering. The dataset is constructed from existing document VQA and document understanding datasets. From these source datasets, we use Docling to extract document layout elements and their semantic types, and **MAPPET** to obtain the bounding box of the evidence region supporting the answer. As part of **MAPPET**, during dataset construction, Qwen2.5-VL-7B-Instruct is used for answer abstraction and as the MLLM underlying the attribution step.
 
 Each item is composed as follows: