Benchmark Data for Model Training
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/hpcaitech/ColossalAI
下载链接
链接失效反馈官方服务：
资源简介：
该数据集的基准测试是在8块Nvidia A100（80GB）GPU上进行的，比较了各种并行策略。数据集还包括了与Megatron-LM、Optimus以及其他模型在性能上的对比，并在所提供的表格中详细列出了设置和性能数据。该测试是基于设备数量进行的弱扩展，任务是模型训练的基准测试。
This dataset's benchmark was conducted on 8 Nvidia A100 (80GB) GPUs to compare various parallelization strategies. The dataset also includes performance comparisons against models such as Megatron-LM, Optimus and other models, with detailed setup configurations and performance data listed in the provided tables. This benchmark follows the weak scaling protocol based on the number of deployed devices, with the core evaluation task being model training.
提供机构：
Colossal-AI
