AbstractPhil/cc-prompts-sharded

该数据集是Conceptual Captions数据集的一个分片版本，专门为Qwen任务转换流程准备。原始数据经过去重处理，仅保留至少4个词的标题，并分割为三个大致相等的分片，以及一个long分片，用于包含超过50个词的长标题（保留供后续高容量模型处理）。每个数据行包括id（在CC流中的位置，用8位数字零填充）、caption（文本标题）和n_words（词数）。id在多次运行中保持稳定，用于下游推理中的恢复跟踪。数据集适用于文本生成任务，语言为英语，大小在100万到1000万样本之间，许可证为CC-BY-4.0。This dataset is a sharded version of the Conceptual Captions dataset, prepared for the Qwen task-conversion pipeline. The original data has been deduplicated, filtered to captions with a minimum of 4 words, and split into three roughly-equal shards plus a long shard for captions over 50 words (reserved for later high-capacity model processing). Each row contains an id (the position in the CC stream, zero-padded to 8 digits), a caption (text), and n_words (word count). The id is stable across re-runs and used for resume tracking in downstream inference. The dataset is intended for text-generation tasks, in English, with a size between 1M and 10M samples, licensed under CC-BY-4.0.