VisInContext

Arxiv

• VisInContext is a easy way to increase the in-context text length in Multi-modality Learning.
• This work is also complement with existing works to increase in-context text length like FlashAttn, Memory Transformer.

Install

pip install -r requirement.txt

For H100 GPUS, run the following dependencies:

pip install -r requirements_h100.txt

Dataset Preparation

See DATASET.md.

Pre-training

See PRETRAIN.md.

Few-shot Evaluation

See Evaluation.md

Citation

If you find our work helps, please consider cite the following work

@article{wang2024visincontext,
        title={Leveraging Visual Tokens for Extended Text Contexts in Multi-Modal Learning},
        author={Wang, Alex Jinpeng and Li, Linjie and Lin, Yiqi and Li, Min  and Wang, Lijuan and Shou, Mike Zheng},
        journal={NeurIPS},
        year={2024}
    }

Contact

Email: awinyimgprocess at gmail dot com

Acknowledgement

Thanks for these good works. Open-flamingo, Open-CLIP and WebDataset.