deqing/addition_dataset

该数据集是一个加法问题集合，格式为{a} + {b} = {c}，其中包含多个子集，如test（5K个评估示例，操作数至少为2位数）、1BT（约85M训练示例，约10亿令牌）、10BT（约850M训练示例，约100亿令牌）和3MT-3digit（详尽单令牌加法，覆盖0到999的所有a和b组合，且a+b≤999）。数据集进行了去重处理，排除了交换性重复项（例如，如果a+b=c存在，则排除b+a=c），并确保测试集对不泄露到训练集中，适用于自然语言处理和机器学习模型的训练与评估。

Addition problems in the format {a} + {b} = {c}. The dataset includes subsets such as test (5K held-out evaluation examples with operands >= 10, i.e., at least 2 digits), 1BT (~85M training examples, ~1 billion tokens under Llama-3 tokenizer), 10BT (~850M training examples, ~10 billion tokens), and 3MT-3digit (exhaustive single-token addition covering all (a, b) with a, b in [0, 999] and a+b ≤ 999). Deduplication is applied to exclude commutative duplicates (e.g., if a+b=c exists, b+a=c is excluded) and prevent test set leakage into training splits, suitable for NLP and machine learning model training and evaluation.