YigeLi/RLAIF-V-Dataset

RLAIF-V-Dataset是一个大规模多模态反馈数据集，提供了83,132个偏好对，指令来自多种数据集，包括MSCOCO、ShareGPT-4V、MovieNet、Google Landmark v2、VQA v2、OKVQA和TextVQA。该数据集旨在通过高质量反馈提升多模态大语言模型（MLLM）的可信度，并已应用于MiniCPM-V和RLAIF-V等模型。数据集包含图像、问题、选择和拒绝的响应，以及数据来源的元信息。

RLAIF-V-Dataset is a large-scale multimodal feedback dataset providing 83,132 preference pairs, with instructions collected from diverse datasets including MSCOCO, ShareGPT-4V, MovieNet, Google Landmark v2, VQA v2, OKVQA, and TextVQA. The dataset aims to enhance the trustworthiness of multimodal large language models (MLLMs) through high-quality feedback and has been applied in models like MiniCPM-V and RLAIF-V. It features images, questions, chosen and rejected responses, and metadata about the datas origin.