rspiocbis/MambaInLlama_0_75-best_of_n-PRM-3464107
收藏Hugging Face2025-02-11 更新2025-02-15 收录
下载链接:
https://hf-mirror.com/datasets/rspiocbis/MambaInLlama_0_75-best_of_n-PRM-3464107
下载链接
链接失效反馈官方服务:
资源简介:
这是一个数学问题数据集,包含问题的描述(problem)、解决方案(solution)、答案(answer)、学科(subject)、难度级别(level)、唯一标识符(unique_id)、生成完成的文本(completions)、分数(scores)、预测结果(pred)、预测使用的token数量(completion_tokens)、聚合分数(agg_scores)以及不同数量token下的加权预测(pred_weighted)、多数投票预测(pred_maj)和天真预测(pred_naive)。数据集分为训练集(train),训练集包含500个样本,大小为148078595字节。此外,还有一个评估配置,包含评估相关的指标如准确率(acc_naive、acc_weighted、acc_maj)和样本数量(n)。
This is a math problem dataset containing problem descriptions (problem), solutions (solution), answers (answer), subjects (subject), difficulty levels (level), unique identifiers (unique_id), generated completion texts (completions), scores (scores), prediction results (pred), number of tokens used in prediction (completion_tokens), aggregated scores (agg_scores), and weighted predictions (pred_weighted), majority vote predictions (pred_maj), and naive predictions (pred_naive) at different token counts. The dataset is split into a training set (train), which contains 500 samples and is 148078595 bytes in size. In addition, there is an evaluation configuration that includes evaluation metrics such as accuracy (acc_naive, acc_weighted, acc_maj) and the number of samples (n).
提供机构:
rspiocbis



