lime-nlp/DeepScaleR_Difficulty
收藏Hugging Face2025-04-10 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/lime-nlp/DeepScaleR_Difficulty
下载链接
链接失效反馈官方服务：
资源简介：
DeepScaleR数据集是一个经过精心策划的推理密集型问题数据集，包含40,000个问题，用于训练和评估基于强化学习的大语言模型。该数据集使用Qwen 2.5-MATH-7B模型进行了难度评分，每个问题尝试128次，根据成功次数计算难度分数。此评分方法旨在平衡不同模型的性能评估。
The DeepScaleR dataset is a curated collection of 40,000 reasoning-intensive problems used for training and evaluating reinforcement learning-based large language models. The dataset is annotated with difficulty scores using the Qwen 2.5-MATH-7B model, with each problem attempted 128 times to compute the difficulty score based on the number of successes.
提供机构：
lime-nlp
