Authors: DAMO NLP Team

Affiliation: Alibaba

Description: 1. Pre-train based on StructBERT
2. Adapted model for MRC by using the DocVQA dataset
3. Ensemble with 30 models and post-process for the final submission.