qfq/eidata_fidelity_cot_improvement_20241025_230429_iter1_fidelity
收藏资源简介:
--- dataset_info: features: - name: doc_id dtype: int64 - name: doc struct: - name: orig_problem dtype: string - name: orig_solution dtype: string - name: orig_answer dtype: string - name: thinking_trajectory sequence: string - name: golden_thinking_trajectory sequence: string - name: model_solution dtype: string - name: old_trajectory sequence: string - name: labeled_trajectory sequence: string - name: text dtype: string - name: problem dtype: string - name: solution dtype: string - name: answer dtype: string - name: target dtype: string - name: arguments struct: - name: gen_args_0 struct: - name: arg_0 dtype: string - name: arg_1 struct: - name: until sequence: string - name: do_sample dtype: bool - name: temperature dtype: float64 - name: max_gen_toks dtype: int64 - name: resps sequence: sequence: string - name: filtered_resps sequence: string - name: doc_hash dtype: string - name: prompt_hash dtype: string - name: target_hash dtype: string - name: exact_match dtype: int64 - name: orig_problem dtype: string - name: orig_solution dtype: string - name: orig_answer dtype: string - name: thinking_trajectory sequence: string - name: golden_thinking_trajectory sequence: string - name: model_solution dtype: string - name: old_trajectory sequence: string - name: labeled_trajectory sequence: string - name: text dtype: string - name: problem dtype: string - name: solution dtype: string - name: answer dtype: string splits: - name: train num_bytes: 696711 num_examples: 35 - name: test num_bytes: 31644 num_examples: 2 download_size: 437447 dataset_size: 728355 configs: - config_name: default data_files: - split: train path: data/train-* - split: test path: data/test-* ---
数据集信息: 特征列表: - 文档ID(doc_id):数据类型为64位整数(int64) - 文档(doc):结构体类型,包含以下子字段: - 原始题目(orig_problem):字符串类型 - 原始解答(orig_solution):字符串类型 - 原始答案(orig_answer):字符串类型 - 思考轨迹(thinking_trajectory):字符串序列 - 标准思考轨迹(golden_thinking_trajectory):字符串序列 - 模型生成解答(model_solution):字符串类型 - 旧轨迹(old_trajectory):字符串序列 - 标注轨迹(labeled_trajectory):字符串序列 - 文本(text):字符串类型 - 题目(problem):字符串类型 - 解答(solution):字符串类型 - 答案(answer):字符串类型 - 目标标签(target):字符串类型 - 参数集(arguments):结构体类型,包含生成参数组0(gen_args_0): - 参数0(arg_0):字符串类型 - 参数1(arg_1):结构体类型,包含以下子字段: - 终止触发序列(until):字符串序列 - 采样开关(do_sample):布尔类型 - 温度系数(temperature):64位浮点数 - 最大生成令牌(Token)数(max_gen_toks):64位整数(int64) - 模型响应(resps):二维字符串序列 - 过滤后响应(filtered_resps):字符串序列 - 文档哈希值(doc_hash):字符串类型 - 提示词哈希值(prompt_hash):字符串类型 - 目标标签哈希值(target_hash):字符串类型 - 精确匹配指标(exact_match):64位整数(int64) 后续重复出现的orig_problem、orig_solution、orig_answer、thinking_trajectory、golden_thinking_trajectory、model_solution、old_trajectory、labeled_trajectory、text、problem、solution、answer字段,均与doc结构体下的同名子字段完全一致,不再另行赘述。 数据集拆分: - 训练集(train):占用字节数696711,共35个样本 - 测试集(test):占用字节数31644,共2个样本 整体数据集参数:下载大小437447,总数据集大小728355 配置项: - 默认配置(default):数据文件映射关系如下 - 训练集拆分:对应路径data/train-* - 测试集拆分:对应路径data/test-*



