QQMM

Release

[2013-12-28]：Model and inference released.

Preparation

Because QQMM use lora tuning the LLM. Before evaluation, you should download the base LLM model:
vicuna-13b-v1.5
QQMM uses visual encoder. Please download from openai/clip-vit-large-patch14-336

Evaluation

Before evaluation, please download the MME benchmark files MME into mme_bench directory.
Then run the cmd:

sh eval/eval_mme.sh

MME Benchmark

QQMM achieved xxx points, which was topx on MME benchmark at 2023-12-28.

Acknowledgments

LLaVA: the codebase we built upon. Thanks for their wonderful work.
Vicuna: the amazing open-sourced large language model!