LehongWu/verl-lt-merged_S6R3_v3_1_0521-gem3f_med-0524_s128tj-replace_cntxt-sft_prior_impl_a256tj
收藏资源简介:
该数据集是一个多模态数据集,包含图像和文本提示(prompt)字段,其中文本提示具有内容和角色属性。数据集还包含奖励模型(reward_model)字段,涉及真实值(ground_truth)和风格(style),以及额外信息(extra_info)字段,如答案(answer)、完成内容(completion)、思考(think)、唯一标识符(uuid)、目标(goal)、任务特定提示(task_specific_prompt)和先前指令(previous_instruction)。数据来源(data_source)、能力(ability)和分割(split)字段提供了元数据。数据集分为训练集(11,785个示例)和测试集(775个示例),总大小约为137MB,可能用于AI模型训练或评估,特别是在多模态和强化学习相关任务中。
This dataset is a multimodal dataset containing image and text prompt fields, where text prompts include content and role attributes. It also includes a reward model field with ground truth and style components, as well as an extra info field with elements such as answer, completion, think, uuid, goal, task-specific prompt, and previous instruction. Metadata is provided through data source, ability, and split fields. The dataset is divided into a training set (11,785 examples) and a test set (775 examples), with a total size of approximately 137MB, likely intended for AI model training or evaluation, particularly in multimodal and reinforcement learning-related tasks.




