https://github.com/salesforce/LAVIS/tree/main/lavis/models/blip2_models LAVIS/lavis/models/blip2_models at main · salesforce/LAVISLAVIS - A One-stop Library for Language-Vision Intelligence - salesforce/LAVISgithub.com Introduction기존 Vision-language pretraining (VLP) 연구의 한계 => end-to-end fashion으로 큰 규모의 모델과 데이터셋을 학습하려다보니 computational cost가 무척 큼일반적이고 compute-efficient한 VLP method 제시, pre-trained..