KletterMix/KletterMix-12B

该数据集是一个德语文本数据集，作为NeurIPS提交的匿名子集发布，用于评审期间的评估、检查和有限的下游实验。数据以分片JSONL文件形式存储，包含100个分片文件，总行数16,826,559，总大小约43.56 GiB。每行数据包括文本内容、聚类ID（整数分配）、令牌计数和代理分数（质量或相关性评分）。数据集旨在支持对聚类结构和分数分布的分析，以及在全集代表性子集上的受控实验。需要注意的是，该子集不代表最终完整版本的全部范围，完整数据集将在相关论文被接受后发布。数据来源基于nvidia/Nemotron-ClimbMix上游数据集。This repository contains an anonymous subset release accompanying a submission to the NeurIPS Datasets and Benchmarks track. The data consists of German-language text examples stored as sharded JSONL files. Each row contains the text itself together with a cluster assignment, a token-count field, and a proxy score. This subset is intended to support evaluation, inspection, and limited downstream experimentation during review. The dataset is split across 100 shard files, with approximately 16,826,559 rows and a total size of about 43.56 GiB. It is designed for analysis of cluster structure and score distributions, and controlled experiments on a representative subset of the full collection. The subset does not represent the complete scope of the final release, and the full dataset will be made available upon acceptance of the associated paper, with the original source being nvidia/Nemotron-ClimbMix.