BidirLM/natcap

BidirLM x Natcap数据集是BidirLM-Omni集合的一部分，用于训练BidirLM-Omni模型。它是一个特定模态的子样本，包含图像和文本对，其中positive特征为图像，anchor特征为字符串（可能表示文本锚点）。数据集有三个配置版本：100k、200k和330k，分别对应不同数量的训练样本（例如，100k版本有100,000个训练示例和500个测试示例），适用于对比学习或多模态任务。该数据集源自更广泛的1.8M样本全模态语料库，但本版本仅聚焦于图像-文本模态。

The BidirLM x Natcap dataset is part of the BidirLM-Omni collection, used for training BidirLM-Omni models. It is a specific modality sub-sample containing image-text pairs, with positive features as images and anchor features as strings (likely representing text anchors). The dataset offers three configurations: 100k, 200k, and 330k, corresponding to different numbers of training samples (e.g., the 100k version includes 100,000 training examples and 500 test examples), suitable for contrastive learning or multimodal tasks. It is derived from a broader 1.8M-sample omnimodal corpus but focuses solely on the image-text modality in this version.