---
license: apache-2.0
---


# [NeurIPS'24]Q-VLM: Post-training Quantization for Large Vision-Language Models

*Efficient and accurate memory saving method towards W4A4 large multi-modal models.* [[Paper](https://arxiv.org/abs/2410.08119)][[Code](https://github.com/ChangyuanWang17/QVLM)]

> Q-VLM: Post-training Quantization for Large Vision-Language Models  
> [Changyuan Wang](https://changyuanwang17.github.io), [Ziwei Wang](https://ziweiwangthu.github.io), [Xiuwei Xu](https://xuxw98.github.io/), [Yansong Tang](https://andytang15.github.io), [Jie Zhou](https://scholar.google.com/citations?user=6a79aPwAAAAJ&hl=en&authuser=1), [Jiwen Lu](http://ivg.au.tsinghua.edu.cn/Jiwen_Lu/)

## Finetuning LLaVA Model on ScienceQA Dataset

Thanks for LLaVA (https://github.com/haotian-liu/LLaVA) for the amazing open-source model!

We combined the LLaVA-7B-v1.1 model ([LLaVA-7B-v1.1](https://huggingface.co/liuhaotian/LLaVA-Lightning-7B-delta-v1-1)) and the projector from LLaVA-7B-v1.3 ([LLaVA-7B-v1.3 projector](https://huggingface.co/liuhaotian/llava-pretrain-vicuna-7b-v1.3/tree/main)) and finetuned the model on the ScienceQA dataset. This model is used to test the effectiveness of our quantization method on the ScienceQA dataset.