Submitted by Zhuchenyang Liu 4 NanoVDR: Distilling a 2B Vision-Language Retriever into a 70M Text-Only Encoder for Visual Document Retrieval NanoVDR 2