jihuny/tldr_1.4b_10k_gopt_policy_norm_penul_last_1e6_iter0
收藏Hugging Face2025-11-03 更新2025-11-15 收录
下载链接：
https://hf-mirror.com/datasets/jihuny/tldr_1.4b_10k_gopt_policy_norm_penul_last_1e6_iter0
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了提示文本(prompt)、索引(prompt_idx)、选中的文本(chosen)、被拒绝的文本(rejected)及其索引和分数等信息。此外，还包含了选中和拒绝文本的完成原因、嵌入向量的差异、去除嵌入向量差异后的分数和排名等字段。数据集分为训练集(train)，共有10000个示例，大小为182,913,901字节。
The dataset includes fields such as prompt, prompt_idx, chosen, rejected, and their indices and scores. It also contains information like the finish reasons for chosen and rejected texts, the difference in embedding vectors, the score after removing the embedding vector difference, and the rank. The dataset is split into a training set (train) with 10,000 examples and a size of 182,913,901 bytes.
提供机构：
jihuny
