ChrisHayduk/nanofold-public
收藏资源简介:
NanoFold Public是nanoFold蛋白质折叠基准的公共训练/验证部分。它打包了一个紧凑、固定、可审计的OpenProteinSet/OpenFold衍生的蛋白质结构训练数据子集,用于在数据高效的折叠模型上进行快速迭代。数据集包含10,000条训练链和1,000条公共验证链,每条链代表一个蛋白质链。原始处理的.npz张量被展开为Hugging Face Dataset列,以便用户可以直接使用typed arrays/lists加载数据。该数据集旨在用于训练和评估较小的蛋白质折叠模型,测试数据高效的架构和目标,以及在没有全规模数据需求的情况下原型化AlphaFold风格的几何学习。它不打算取代全规模的OpenProteinSet/OpenFold训练数据,而是一个故意受限的基准切片。
NanoFold Public is the public train/validation portion of the nanoFold protein-folding benchmark. It packages a compact, fixed, auditable subset of OpenProteinSet/OpenFold-derived protein structure training data for fast iteration on data-efficient folding models. The dataset has 10,000 train chains and 1,000 public validation chains. Each row is one protein chain. The original processed .npz tensors are unrolled into Hugging Face Dataset columns so users can load the data with datasets.load_dataset and work directly with typed arrays/lists. This dataset is intended for training and evaluating smaller protein-folding models, testing data-efficient architectures and objectives, and prototyping AlphaFold-style geometry learning without full-scale data requirements. It is not intended to replace full-scale OpenProteinSet/OpenFold training data. It is a deliberately constrained benchmark slice.




