Skip to content

M2RAG

A benchmark for multimodal retrieval generation

Visit official site

What M2RAG is

The codebase is described as a benchmark for retrieval-augmented generation in multimodal contexts, built to test multimodal language models.

What you can do with it

  • Evaluate a model on image captioning
  • Run fact verification and image reranking
  • Use MM-RAIT instruction tuning

Getting started

  • Clone the codebase from GitHub
  • Install packages from requirements.txt
  • Place the dataset in the data folder

Pros & cons

✓ What we like

  • Full datasets and code are provided
  • Zero-shot and fine-tuned settings are supported

! What to watch out for

  • The codebase is framed as a research benchmark
  • Evaluation metrics are tied to generation tasks

FAQ

Which tasks are covered?

The material lists image captioning, multimodal QA, fact verification and image reranking.

What models are named?

The material names MiniCPM-V 2.6 and Qwen2-VL as compatible pre-trained models.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review