DDSC/nordic-embedding-training-data
收藏Hugging Face2025-12-13 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/DDSC/nordic-embedding-training-data
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是为了预训练或后训练丹麦语的嵌入模型而设计，适用于文本相似度任务。它包含了968k个样本，每个样本包括查询、正样本、硬负样本、样本语言（丹麦语、挪威语或瑞典语）、任务类型、任务指令、原始提示和原始响应。数据集使用InfoNCE损失函数结构，适用于检索和单元三元组任务。
This dataset is designed for pre-training or post-training embedding models for Danish on text similarity tasks. It consists of 968k samples, each including a query, a positive sample, a hard negative sample, the sample language (Danish, Norwegian, or Swedish), the task type, the task instruction, the original prompt, and the original response. The dataset is structured using InfoNCE loss function and is suitable for retrieval and unit-triplet tasks.
提供机构：
DDSC
