RLHF-And-Friends/PPO-vs-CoPPO
收藏Hugging Face2025-03-20 更新2025-04-26 收录
下载链接：
https://hf-mirror.com/datasets/RLHF-And-Friends/PPO-vs-CoPPO
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了两个模型对同一提示的响应数据。具体包括一个提示列（prompt）和两个模型响应列，分别对应TLDR-Mistral-7B-SmallSFT-PPO模型和TLDR-Mistral-7B-SmallSFT-CoPPO模型的响应。数据来源于RLHF-And-Friends/tldr-ppo数据集。此外，还提供了TLDR-Mistral-7B-SmallSFT-CoPPO模型相对于TLDR-Mistral-7B-SmallSFT-PPO模型的胜率为0.707，该胜率是基于gpt-4o-mini模型的意见得出的。
This dataset includes the responses from two models to the same prompt. It consists of a prompt column and two response columns corresponding to the responses from the TLDR-Mistral-7B-SmallSFT-PPO model and the TLDR-Mistral-7B-SmallSFT-CoPPO model, respectively. The data is sourced from the RLHF-And-Friends/tldr-ppo dataset. Additionally, the winrate of TLDR-Mistral-7B-SmallSFT-CoPPO over TLDR-Mistral-7B-SmallSFT-PPO is 0.707, based on the opinion of the gpt-4o-mini model.
提供机构：
RLHF-And-Friends
