elandy/cje-chatbot-arena

CJE Chatbot Arena数据集来自Causal Judge Evaluation在Chatbot Arena数据上的实验。数据集包含主数据集文件cje_dataset.jsonl（4,961个提示）、原始Chatbot Arena提示文件prompts.jsonl、模型响应文件夹responses/和token logprobs文件夹logprobs/。评估了5种系统提示变体，包括无系统提示的base、clone、parallel_universe_prompt、premium和unhelpful（负控制）。The CJE Chatbot Arena Dataset is from the Causal Judge Evaluation experiments on Chatbot Arena data. The dataset includes the main dataset file cje_dataset.jsonl (4,961 prompts), the original Chatbot Arena prompts file prompts.jsonl, the model responses folder responses/, and the token logprobs folder logprobs/. Five system prompt variants were evaluated, including base (no system prompt), clone, parallel_universe_prompt, premium, and unhelpful (negative control).