allenai/big-reasoning-traces
收藏Hugging Face2025-04-01 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/allenai/big-reasoning-traces
下载链接
链接失效反馈官方服务：
资源简介：
DeepSeek是一个大规模的推理轨迹数据集，适用于中期训练/退火前的实验。数据集包含大约25亿个标记，使用了OLMo 2分词器。它由多个来源的数据集组合而成，包括GeneralThought-430K、OpenThoughts-114k和OpenR1-Math-220k，这些来源提供了大量的示例和标记。数据集分为训练集，其中DeepSeek配置的训练集有676665个示例，DeepSeek_debug配置的训练集有300个示例。
DeepSeek is a large-scale reasoning trace dataset designed for experiments with midtraining/annealing. The dataset contains approximately 2.5 billion tokens and is tokenized using the OLMo 2 tokenizer. It is compiled from multiple source datasets including GeneralThought-430K, OpenThoughts-114k, and OpenR1-Math-220k, which together provide a substantial number of examples and tokens. The dataset is split into a training set, with the DeepSeek configuration having 676,665 examples and the DeepSeek_debug configuration having 300 examples.
提供机构：
allenai
