Share4oReasoning/dpo_data
收藏Hugging Face2024-12-27 更新2025-04-26 收录
下载链接：
https://hf-mirror.com/datasets/Share4oReasoning/dpo_data
下载链接
链接失效反馈官方服务：
资源简介：
ShareGPT4oReasoning训练数据集（DPO）是一个用于视觉问答和图像文本到文本任务的数据集。它包含了模型生成的CoT（Chain-of-Thought）判断结果奖励，以及用于SFT（Soft Prompt Tuning）的压缩图像数据。数据集支持英文，大小小于1M。数据集包含了多种图像数据，如ai2d_images、aokvqa、coco等，以及对应的训练指令。
ShareGPT4oReasoning training dataset (DPO) is a dataset for question answering and image-text-to-text tasks. It contains model generated CoT (Chain-of-Thought) outcome reward judgments and zipped image data for SFT (Soft Prompt Tuning). The dataset supports English language and is smaller than 1M in size. It includes various image data such as ai2d_images, aokvqa, coco, etc., along with corresponding training instructions.
提供机构：
Share4oReasoning
