marin-community/synth-science-markup-tree-ppl

该数据集是一个多任务、多格式的文本补全数据集，包含多个配置，如BibTeX条目补全、HTML实体属性闭合、LaTeX引用闭合、FASTA序列记录、GenBank特征记录、MathML嵌套树、PDB原子记录、SDF记录闭合、SMILES公式补全、SVG组路径闭合和XML标签闭合等。每个配置都针对特定结构化文本格式，提供输入和目标字段，以及相关元数据，用于评估模型在生成或补全结构化文本时的性能。数据集仅包含验证分割，每个配置有1000个示例，适用于自然语言处理任务中的序列生成和格式理解。

This dataset is a multi-task, multi-format text completion dataset comprising various configurations such as BibTeX entry completion, HTML entity attribute closure, LaTeX reference closure, FASTA sequence record, GenBank feature record, MathML nested tree, PDB atom records, SDF record closure, SMILES formula completion, SVG group path closure, and XML tag closure. Each configuration focuses on a specific structured text format, providing input and target fields along with relevant metadata, designed to evaluate model performance in generating or completing structured text. The dataset includes only a validation split with 1000 examples per configuration, suitable for sequence generation and format understanding tasks in natural language processing.