ServiceNow-AI-R1-Distill-SFT
收藏资源简介:
SLAM lab发布的R1-Distill-SFT数据集是通过DeepSeek-R1-32b模型蒸馏生成的,使用了Numina-math和Tulu工具,并且每个提示只采样一个响应。数据集分为多个版本(v0和v1),每个版本包含不同的特征和样本数量。v0版本包含171,647个样本,v1版本包含1,679,162个样本。数据集的特征包括id、reannotated_assistant_content、problem、source、solution等。数据集的发布计划包括种子集、未过滤/未验证数据集、过滤和验证版本以及SFT模型的发布。
The R1-Distill-SFT dataset released by SLAM Lab is distilled from the DeepSeek-R1-32b model, utilizing the Numina-math and Tulu tools, with exactly one response sampled per prompt. The dataset has multiple versions (v0 and v1), each with distinct features and sample counts. The v0 version contains 171,647 samples, while the v1 version contains 1,679,162 samples. The features of the dataset include id, reannotated_assistant_content, problem, source, solution, etc. The release plan for this dataset covers seed datasets, unfiltered/unvalidated datasets, filtered and validated versions, as well as the release of SFT models.




