THUDM/LongReward-10k
收藏官方服务：
资源简介：
LongReward-10k数据集包含10,000个长上下文问答实例，涵盖英文和中文两种语言，每个实例最长可达64,000个单词。数据集分为三个部分：sft部分包含监督微调数据，dpo_glm4_9b和dpo_llama3.1_8b部分为长上下文偏好数据集，用于训练不同的DPO模型。
The LongReward-10k dataset contains 10,000 long-context QA instances in both English and Chinese, with each instance up to 64,000 words in length. The dataset is divided into three parts: the sft part includes supervised fine-tuning data, while the dpo_glm4_9b and dpo_llama3.1_8b parts are long-context preference datasets used for training different DPO models.
提供机构：
THUDM
