KaLM-Embedding/KaLM-embedding-finetuning-data-spanish

KaLM-embedding-finetuning-data-spanish是一个用于嵌入模型微调的西班牙语数据集，改编自上游数据集。该数据集保持了与上游版本相同的面向训练的三元组/列表结构，并组织为多个可独立加载或组合用于大规模嵌入训练的parquet支持的子集。数据集包含85个子集，共162个parquet分片，本地磁盘使用约30G。每个样本遵循相同的嵌入微调格式：query（字符串，每个样本一个查询），pos（字符串列表，通常包含一个正例），neg（字符串列表，通常包含七个负例）。数据集适用于检索式嵌入微调、查询-文档对比学习、句子相似性/STS风格训练以及西班牙语数据的多语言或跨语言嵌入适应。

KaLM-embedding-finetuning-data-spanish is a Spanish finetuning dataset for embedding models, adapted from the upstream dataset. It maintains the same training-oriented triplet/list structure as the upstream release and is organized into multiple parquet-backed subsets that can be loaded independently or combined for large-scale embedding training. The dataset includes 85 subsets with 162 parquet shards, occupying about 30G of local disk space. Each sample follows the same embedding finetuning format: query (string, one query per sample), pos (list[string], usually containing one positive example), neg (list[string], usually containing seven negative examples). The dataset is suitable for retrieval-style embedding finetuning, query-document contrastive learning, sentence similarity / STS-style training, and multilingual or cross-lingual embedding adaptation with Spanish data.