Is there any state of art methods used in the insuranceQA (v2)?

The original datasets have two version: version 1 and version 2. 
I have check some papers, but they all experiments on version 1 dataset and the state of art model achieved 0.77, is there any paper experiments on version 2 dataset?