wtd/collect-Omni-MATH-qwen25-7b

该数据集是一个用于强化学习或交互式任务的数据集，包含多个特征字段，如episode_id（剧集ID）、timestep（时间步）、actions（动作）、text_actions（文本动作）、raw_text_actions（原始文本动作）、judge_feedbacks（判断反馈）、plans（计划）、rewards（奖励）、terms（终止标志）、truncs（截断标志）等。此外，还包括image_paths（图像路径）和state_dumps（状态转储）等视觉和状态数据字段，以及obs_problem（观察问题）、obs_solution（观察解决方案）、obs_answer（观察答案）等与问题解决相关的字段。数据集还记录了llm_backend（大型语言模型后端）、llm_model_id（大型语言模型ID）等信息，暗示可能涉及基于大型语言模型的交互。数据集分为train分割，包含355,495个示例，总大小约为10.2 GB，下载大小约为1.8 GB。This dataset is designed for reinforcement learning or interactive tasks, featuring multiple fields such as episode_id, timestep, actions, text_actions, raw_text_actions, judge_feedbacks, plans, rewards, terms, truncs, and others. It also includes visual and state data fields like image_paths and state_dumps, as well as problem-solving related fields like obs_problem, obs_solution, and obs_answer. The dataset records information such as llm_backend and llm_model_id, suggesting potential involvement with large language model-based interactions. It is split into a train partition with 355,495 examples, total size approximately 10.2 GB, and download size approximately 1.8 GB.