AvaXiao/ReToken-InternVL3.5-8B

ReToken: One Token to Improve Vision–Language Models for Visual Retrieval

This repository contains the model from the paper ReToken: One Token to Improve Vision-Language Models for Visual Retrieval.

ReToken is a single learnable embedding trained as an explicit retrieval target that selects a sparse set of query-relevant visual tokens from a pre-filled visual KV cache.

Code: https://github.com/avaxiao/ReToken