princeton-nlp/llama3-ultrafeedback-armorm

该数据集名为llama3-ultrafeedback-armorm，用于训练princeton-nlp/Llama-3-Instruct-8B-SimPO-v0.2模型。数据集包含约60k训练样本和2k测试样本，遵循HuggingFaceH4/ultrafeedback_binarized的原始分割。数据集的结构包括prompt_id、prompt、all_generated_responses、all_rm_scores、chosen和rejected等字段。chosen和rejected字段分别包含content和role子字段。数据集的许可证为MIT。

This dataset is used to train the [princeton-nlp/Llama-3-Instruct-8B-SimPO-v0.2](https://huggingface.co/princeton-nlp/Llama-3-Instruct-8B-SimPO-v0.2) model. The dataset includes features such as prompt_id, prompt, all_generated_responses, all_rm_scores, chosen, and rejected. It contains both training and testing splits, with detailed information on the number of examples and bytes for each split. The dataset is licensed under the MIT license. The README also provides an example entry to illustrate the structure of the data.