Salesforce/CogAlign

CogAlign是一个为视觉语言模型（VLMs）设计的后训练策略，旨在增强它们的视觉算术能力。这个数据集包含64000个合成示例，用于支持这种后训练过程。每个示例包括一个视觉输入、一个提示比较特定属性的查询、一个与视觉输入一致的积极回应和一个与之矛盾的消极回应。通过训练VLMs使用CogAlign，可以在依赖于视觉算术的下游任务中提高性能。

CogAlign is a post-training strategy for Vision Language Models (VLMs) designed to enhance their visual arithmetic capabilities. This dataset contains 64,000 synthetic examples intended to facilitate the post-training process. Each example includes a visual input, a query prompting comparison of a specific property, a positive response consistent with the visual input, and a negative response that contradicts it. Training VLMs with CogAlign leads to performance improvements in downstream tasks involving visual arithmetic.