JUNJIE99/MegaPairs
收藏官方服务：
资源简介：
MegaPairs是一个包含超过2600万三元组的大规模数据集，用于通用多模态检索任务。通过开放的领域图像创建异质KNN三元组，支持训练多模态检索模型BGE-VL，该模型在多个检索基准测试中表现出优异性能。
MegaPairs is a large-scale dataset containing over 26 million triplets for universal multimodal retrieval tasks. It creates heterogeneous KNN triplets from open-domain images to support the training of the multimodal retrieval model BGE-VL, which exhibits excellent performance on several retrieval benchmarks.
提供机构：
JUNJIE99
