DataComp

DataComp是一个机器学习基准测试平台，专注于固定模型条件下寻找最优训练数据。该平台提供多种任务设置，包括CLIP图像文本预训练、语言建模和推理链等，用户可从大规模数据集中选择最佳子集来训练模型，并通过下游任务评估训练效果。DataComp is a machine learning benchmark platform dedicated to identifying optimal training data under fixed model conditions. This platform provides diverse task settings, including CLIP-based image-text pre-training, language modeling, chain-of-thought reasoning, and more. Users can select optimal subsets from large-scale datasets to train models, and evaluate the performance of the trained models via downstream tasks.