The-CoLab/multilingual-textarena-Nim-v0-train
收藏资源简介:
该数据集包含语言条件化的TextArena轨迹数据,用于文本生成任务。每个数据集配置对应不同的模型、实验组或源文件夹,具体包括gemma4-e4b-it、qwen3-4b和ministral3-3b-instruct配置。数据以parquet文件格式存储,涵盖训练分割,包含行级轨迹记录,常见列有game_idx、step_idx、env_id、pid、policy_id、model_name、observation、raw_action、raw_action_len、action、reward、policy_epoch、step_info、game_info、lang和lang_map等,部分列可能因原始文件而异。嵌套字段如step_info、game_info和lang_map以字符串形式存储以确保兼容性。数据集基于MIT许可证,语言为英语。
This dataset contains language-conditioned TextArena trajectory data for text generation tasks. Each dataset configuration corresponds to a specific model, experimental group, or source folder, with the included configurations being gemma4-e4b-it, qwen3-4b, and ministral3-3b-instruct. The data is stored in Parquet file format, covering the training split, and contains row-level trajectory records. Common columns include game_idx, step_idx, env_id, pid, policy_id, model_name, observation, raw_action, raw_action_len, action, reward, policy_epoch, step_info, game_info, lang, and lang_map, while some columns may vary across the original files. Nested fields such as step_info, game_info, and lang_map are stored as strings to ensure compatibility. This dataset is licensed under the MIT License, with English as its primary language.




