violetxi/Qwen8b-NoThink-Judge-int-qwen8b-multi-ref
收藏资源简介:
该数据集包含1076个训练示例,用于评估语言模型在问题解决任务中的表现。每个示例包括一个问题(problem)、推理轨迹(reasoning_trace)、参考解决方案(reference_solution),以及表示思考是否完成的布尔标志(finish_thinking)。此外,还包含奖励(reward)和分数(score)字段用于量化性能。数据集还提供了错误参考问题(wrong_reference_problem)和错误参考解决方案(wrong_reference_solution),以及三个不同条件下的LLM评估结果:参考解决方案条件下(llm_score_ref、llm_response_ref)、无参考解决方案条件下(llm_score_noref、llm_response_noref)和错误参考解决方案条件下(llm_score_wrongref、llm_response_wrongref)。这些特征支持对语言模型在多种问题解决场景中的分析和比较。
This dataset contains 1076 training examples designed to evaluate the performance of language models on problem-solving tasks. Each example includes a problem, a reasoning trace, a reference solution, and a boolean flag indicating whether thinking is completed (finish_thinking). Additionally, it features reward and score fields to quantify performance. The dataset also provides wrong reference problems and solutions, along with LLM evaluation results under three conditions: with reference solution (llm_score_ref, llm_response_ref), without reference solution (llm_score_noref, llm_response_noref), and with wrong reference solution (llm_score_wrongref, llm_response_wrongref). These features facilitate analysis and comparison of language models across various problem-solving scenarios.




