collinear-ai/R1-Distill-NuminaMath-Curation

NuminaMath数据集是从ServiceNow-AI/R1-Distill-SFT数据集中提取的NuminaMath样本，包含问题、解决方案、来源、R1-Distill-Qwen-32B模型响应、R1-Distill-Qwen-32B消息格式响应以及正确性判断。问题、解决方案和来源字段来自NuminaMath数据集，对应于问题陈述、真实解决方案和问题来源。R1-Distill-Qwen-32B字段包含Qwen 32B模型对问题的响应，R1-Distill-Qwen-32B消息字段是以聊天格式呈现的响应。正确性字段包含由Collinear审校者C1和C2以及GPT-4o对生成响应正确性的判断。

The NuminaMath dataset consists of samples from the ServiceNow-AI/R1-Distill-SFT dataset and includes features for problem, solution, source, responses generated by the Qwen 32B model (R1-Distill-Qwen-32B), chat-formatted responses (R1-Distill-Qwen-32B-messages), and correctness judgments. The problem, solution, and source features are derived from the NuminaMath dataset, corresponding to the problem statement, ground truth solution, and problem source, respectively. The R1-Distill-Qwen-32B feature contains responses to the problem by the Qwen 32B model, and the R1-Distill-Qwen-32B-messages feature presents the responses in a chat format. The correctness feature includes judgments on the correctness of the generated response by Collinear curators C1 and C2, and GPT-4o, provided as a json object where the values for the keys ‘C1’, ‘C2’, and ‘gpt’ are the curator predictions regarding the response correctness.