marin-community/synth-identifier-encoding-ppl

该数据集是一个多配置数据集，设计用于评估模型在多种数据格式和编码任务上的延续或生成能力。它包括10个子任务：Base64编码延续、生物信息学访问号处理、提交哈希生成、数据图像Base64前缀处理、十六进制转储延续、JSON Unicode字节转义、混合大小写符号标识符生成、包名和版本处理、URL查询转义以及UUID构建ID生成。每个子任务配置包含输入字符串、目标字符串、唯一标识符、子集分类、任务标识、随机种子和元数据（如家族、语法、行索引等）。数据集仅提供验证集，每个子任务有1000个示例，用于测试模型在结构化数据生成和编码方面的性能。

This dataset is a multi-configuration dataset designed to evaluate model performance on continuation or generation tasks across various data formats and encodings. It includes 10 sub-tasks: Base64 continuation, bioinformatics accessions, commit hashes, data image Base64 prefixes, hex dump continuation, JSON Unicode byte escapes, mixed-case symbolic identifiers, package names and versions, URL query escaping, and UUID build IDs. Each sub-task configuration features input strings, target strings, unique identifiers, subset classifications, task identifiers, random seeds, and metadata (such as family, grammar, row index, etc.). The dataset provides only validation splits, with 1000 examples per sub-task, aimed at testing model capabilities in structured data generation and encoding tasks.