openbmb/DensingLaw-ScalingBench
收藏Hugging Face2025-08-04 更新2025-08-09 收录
下载链接：
https://hf-mirror.com/datasets/openbmb/DensingLaw-ScalingBench
下载链接
链接失效反馈官方服务：
资源简介：
DensingLaw-ScalingBench数据集旨在帮助研究者更精确地估计大型语言模型（LLMs）的性能缩放规律。该数据集包含为计算条件损失而特别设计的测试实例，这些条件损失是在下游任务上对LLM密度进行评估的一部分。数据集分为两种类型的任务：多项选择题和需要多步骤推理的复杂问题。数据集在Apache 2.0许可下发布。
The DensingLaw-ScalingBench dataset is designed to facilitate more accurate performance scaling law estimation of Large Language Models (LLMs). The dataset contains test instances specifically crafted for calculating conditional loss, which is part of the LLM density evaluation on downstream tasks. It includes two main types of tasks: multiple-choice problems and complex reasoning problems. The dataset is released under the Apache 2.0 license.
提供机构：
openbmb
