tinygsm_fopython_no_obs_depth1to9_traindepth5

该数据集包含多轮对话或问题解答场景的结构化记录，主要特征包括：问题唯一标识(question_hash)、计算深度(computation_depth)、赋值操作次数(num_assignments)、原始轨迹长度(primitive_trace_len)、最终答案(answer)、对话轮次(num_turns)以及完整的对话文本(turns/text)。数据集划分为训练集(6,841,488条)、测试集(760,686条)和OOD测试集(263,037条)三部分，总大小约12.5GB。每个样本包含从问题标识到最终答案的完整交互链条，适用于对话系统训练、问答模型评估等NLP任务，特别包含专门测试分布外泛化能力的独立测试集(test_ood)。This dataset contains structured records of multi-turn dialogue or question answering scenarios. Its main features include: unique question identifier (question_hash), computation depth, number of assignment operations (num_assignments), length of original primitive trace (primitive_trace_len), final answer (answer), number of dialogue turns (num_turns), and complete dialogue text (turns/text). The dataset is divided into three subsets: training set with 6,841,488 samples, test set with 760,686 samples, and out-of-distribution (OOD) test set with 263,037 samples, with a total size of approximately 12.5 GB. Each sample contains the complete interaction chain from the question identifier to the final answer, which is suitable for NLP tasks such as dialogue system training and question answering model evaluation. Specifically, it includes an independent test set (test_ood) dedicated to testing out-of-distribution generalization ability.