rasdani/deepcoder_eval_claude-sonnet-4-20250514

该数据集包含了一系列的例子，每个例子都有唯一的标识符、提示(prompt)、完成(completion)、任务类型(task)、奖励(reward)、生成时间(generation_ms)、评分时间(scoring_ms)、总时间(total_ms)、额外信息(info)、答案(answer)和深度编码器奖励函数(deepcoder_reward_func)。提示和完成都包含内容和角色两个子字段。额外信息中包含了函数名称(fn_name)、子集(subset)、子集索引(subset_idx)以及测试(tests)的输入和输出。数据集分为训练集，大小为126141字节，共有16个示例。默认配置下，训练数据文件以data/train-*的模式存储。

The dataset consists of a series of examples, each with a unique identifier, prompt, completion, task type, reward, generation time (generation_ms), scoring time (scoring_ms), total time (total_ms), additional info (info), answer, and deepcoder reward function (deepcoder_reward_func). Both prompt and completion include sub-fields for content and role. Additional info contains function name (fn_name), subset (subset), subset index (subset_idx), and tests with inputs and outputs. The dataset is split into a training set, which is 126141 bytes in size and contains 16 examples. Under the default configuration, the training data files are stored following the pattern data/train-*.