teamcore/DPO_Pm3B_U0_beta0.25sigmoidEurus_RM_7bbt_noise_flip0.3g

该数据集名为test_run_rebuttal_check，包含多个字段，主要用于评估模型的响应质量。特征字段包括：源文本(source)、指令(instruction)、模型名称(models)、注释(annotations)、评分(rating)、评分理由(rationale)、评分类型(type)、诚实度(honesty)、指令遵循度(instruction_following)、真实性(truthfulness)、批评(critique)、自定义系统提示(custom_system_prompt)、细粒度分数(fine-grained_score)、模型名称(model)、总分数(overall_score)、原则(principle)、响应(response)、正确答案(correct_answers)、错误答案(incorrect_answers)、提示(prompt)、选中(chosen)、拒绝(rejected)、选中分数(chosen_score_Eurus_RM_7b)、拒绝分数(rejected_score_Eurus_RM_7b)、bt概率(bt_prob_Eurus_RM_7b)、响应(response)、生成奖励分数(reward_score_generated)、选择奖励分数(reward_score_chosen)、GPT分数(gpt_score)、GPT反馈(gpt_feedback)。数据集分为default一个split，包含100个示例。

The dataset named test_run_rebuttal_check includes multiple fields primarily used for evaluating the quality of model responses. Features include: source text (source), instruction (instruction), model names (models), annotations (annotations), ratings (rating), rationales (rationale), rating types (type), honesty (honesty), instruction following (instruction_following), truthfulness (truthfulness), critique (critique), custom system prompt (custom_system_prompt), fine-grained scores (fine-grained_score), model names (model), overall scores (overall_score), principles (principle), responses (response), correct answers (correct_answers), incorrect answers (incorrect_answers), prompts (prompt), chosen (chosen), rejected (rejected), chosen scores (chosen_score_Eurus_RM_7b), rejected scores (rejected_score_Eurus_RM_7b), bt probabilities (bt_prob_Eurus_RM_7b), responses (response), generated reward scores (reward_score_generated), chosen reward scores (reward_score_chosen), GPT scores (gpt_score), and GPT feedback (gpt_feedback). The dataset is split into a default split containing 100 examples.