teamcore/DPO_Pm3B_U0_beta0.25dpo_proEurus_RM_7bbt_noise_flip_paper0.6g
收藏资源简介:
该数据集包含了两个配置:test_run_rebuttal_noisefixed和test_run_rebuttal_noisefixed_smallnu。每个配置都有多个特征,包括文本来源、指令、模型列表、完成情况(包括帮助性、诚实度、指令遵循度和真实性等评价标准)、正确和错误答案、提示、选择和拒绝的答案、不同模型的分数、响应、生成和选择的奖励分数、GPT评分和反馈。数据集分为默认split,每个split有1000个示例。
The dataset includes two configurations: test_run_rebuttal_noisefixed and test_run_rebuttal_noisefixed_smallnu. Each configuration has multiple features, including text source, instruction, model list, completions (including ratings for helpfulness, honesty, instruction following, and truthfulness), correct and incorrect answers, prompts, chosen and rejected answers, scores for different models, responses, generated and chosen reward scores, GPT score, and feedback. The dataset is split into a default split, with each split containing 1000 examples.



