LehongWu/verl-lt-collect_6tasks_013789_v3_1_0416-gemini3flash_medium-repeat8_0416_suc100trajs

该数据集是一个多模态数据集，包含图像和文本提示，用于训练或测试奖励模型。特征包括：images（图像列表）、prompt（包含内容和角色的文本列表）、reward_model（包含ground_truth和风格的结构化数据）、extra_info（包含答案、完成内容、思考、UUID、目标、任务特定提示和先前指令的附加信息）、data_source（数据来源）、ability（能力描述）和split（数据集划分）。数据集分为训练集（5455个样本）和测试集（523个样本），总大小约268MB，下载大小约79MB。

This dataset is a multimodal dataset containing images and text prompts, designed for training or testing reward models. Features include: images (list of images), prompt (list of text with content and role), reward_model (structured data with ground_truth and style), extra_info (additional information including answer, completion, think, uuid, goal, task_specific_prompt, and previous_instruction), data_source (data source), ability (ability description), and split (dataset split). The dataset is divided into a training set (5455 examples) and a test set (523 examples), with a total size of approximately 268MB and a download size of approximately 79MB.