open-instruct-openthoughts
收藏资源简介:
该数据集名为OpenThoughts Agent - Open Instruct formatted,是一个经过强化学习(RL)格式化处理的数据集,专门适配于特定版本的open-instruct框架,旨在为智能体训练提供指令遵循数据。每个样本包含多个结构化字段:messages字段是一个列表,其中每条消息包含content(文本内容)和role(角色标识);ground_truth字段提供了预期答案或目标输出;dataset字段标识数据来源;env_config字段是一个结构体,描述了环境配置,包括环境名称(env_name)、镜像(image)和任务ID(task_id);source字段记录了数据原始出处。数据集仅包含一个训练分片(train),共有728个样本,总数据大小约为2.65 MB,适用于指令微调、智能体策略学习或对话系统训练等任务。原始数据来源于OpenThoughts-Agent-v1-RL数据集。
This dataset, named OpenThoughts Agent - Open Instruct formatted, is a reinforcement learning (RL) formatted dataset specifically adapted for a particular version of the open-instruct framework. It is designed to provide instruction-following data for agent training. Each sample includes multiple structured fields: the messages field is a list where each message contains content (text content) and role (role identifier); the ground_truth field provides the expected answer or target output; the dataset field identifies the data source; the env_config field is a structure describing environment configuration, including environment name (env_name), image, and task ID (task_id); and the source field records the original data provenance. The dataset contains only one training shard (train) with 728 samples, totaling approximately 2.65 MB in size. It is suitable for tasks such as instruction fine-tuning, agent policy learning, or dialogue system training. The original data comes from the OpenThoughts-Agent-v1-RL dataset.
数据集名称
OpenThoughts Agent - Open Instruct formatted
数据集来源
- 原始数据集:OpenThoughts-Agent-v1-RL(地址:https://huggingface.co/datasets/open-thoughts/OpenThoughts-Agent-v1-RL)
- 本数据集为原始数据的重新格式化版本,专用于 open-instruct 框架的强化学习场景。
数据特征
- messages:包含对话消息列表,每条消息包含:
content(字符串):消息内容role(字符串):消息角色
- ground_truth(字符串):真实答案或标注
- dataset(字符串):所属数据集名称
- env_config:环境配置,包含:
env_name(字符串):环境名称image(字符串):环境镜像task_id(字符串):任务ID
- source(字符串):数据来源
数据集划分
- 训练集(train):包含 728 个样本,数据大小为 2,653,161 字节
相关资源
- 代码仓库:https://github.com/hamishivi/tmax
- 模型与数据集合:https://huggingface.co/collections/allenai/tmax
- 论文:https://arxiv.org/abs/2606.23321
- 博客:https://wai-org.com/blog/tmax/




