MemLens

MemLens: Benchmarking Multimodal Long-Context Conversational Memory in Vision-Language Models This repository hosts the MemLens dataset only. Evaluation code, model wrappers, and scoring scripts live at github.com/xrenaf/MEMLENS. Overview MemLens is a benchmark for evaluating long-horizon conversational memory in vision-language models. It tests whether models can retrieve, recall, update, and reason… See the full description on the dataset page:

Source: Hugging Face Hub (xiyuRenBill/MEMLENS). Metadata imported from the dataset’s Hub tags.