karmakorma/sasbuddylm-v2
收藏Hugging Face2026-04-22 更新2026-04-26 收录
下载链接：
https://hf-mirror.com/datasets/karmakorma/sasbuddylm-v2
下载链接
链接失效反馈官方服务：
资源简介：
SASBuddyLM v2 是一个用于微调临床SAS代码助手的数据集，包含ChatML格式的规范到程序对。数据集按研究级别划分，包括训练集（102个）、验证集（0个）和测试集（100个），且每个研究仅出现在一个划分中。每条记录以JSONL格式存储，包含角色消息、研究、数据集和划分信息。数据来源于通过GPT-5.4从每个变量的规范代码片段中提取并重写为自包含的SAS程序。
SASBuddyLM v2 is a dataset for fine-tuning a clinical SAS code assistant, containing ChatML-formatted spec-to-program pairs. The dataset is partitioned at the study level, including train (102), validation (0), and test (100) splits, with no study appearing in more than one split. Each record is stored in JSONL format, containing role messages, study, dataset, and split information. The data was distilled from per-variable specification-code fragments aggregated by study and dataset, using GPT-5.4 to rewrite fragments into self-contained SAS programs.
提供机构：
karmakorma
