nomic-ai/cornstack-java-v1

CoRNStack数据集是一个大规模的高质量训练数据集，用于跨多种编程语言的代码检索。该数据集由<查询，正例，负例>三元组组成，用于训练nomic-embed-code、CodeRankEmbed和CodeRankLLM模型。数据集从去重的Stackv2开始构建，通过筛选出高质量的文本-代码对，并进行双重一致性过滤以去除噪声样本。训练过程中采用了新颖的课程式硬负样本挖掘策略。

The CoRNStack Dataset is a large-scale high-quality training dataset specifically designed for code retrieval across multiple programming languages. It consists of `<query, positive, negative>` triplets for training the nomic-embed-code, CodeRankEmbed, and CodeRankLLM models. The dataset is constructed starting from the deduplicated Stackv2, creating high-quality text-code pairs from function docstrings and respective code, and applying dual-consistency filtering to remove noisy examples. A novel curriculum-based hard negative mining strategy is employed during training.