home/categories/framework-internals/ascend-ai-coding-awesome-ascend-skills-torch-npu-comm-test-skill-md
framework-internalsdevelopment
torch-npu-comm-test
通过 PyTorch torch.distributed 接口测试昇腾 NPU 通信算子性能。支持指定任意 tensor shape、dtype，使用 torchrun 启动，贴近真实训练场景的通信算子测试与性能分析。Use for testing collective communication operators (AllReduce, AllGather, ReduceScatter, etc.) with specific tensor shapes via torch.distributed on Ascend NPU.
maintainer
ascend-ai-coding
更新于 3/26/2026
星标
48
分支
30
quick start
Installation and usage
通过 PyTorch torch.distributed 接口测试昇腾 NPU 通信算子性能。支持指定任意 tensor shape、dtype，使用 torchrun 启动，贴近真实训练场景的通信算子测试与性能分析。Use for testing collective communication operators (AllReduce, AllGather, ReduceScatter, etc.) with specific tensor shapes via torch.distributed on Ascend NPU.
安装
$ install --globalskills.sh
使用
安装后，您可以通过在终端运行以下命令来使用此技能：
skills use torch-npu-comm-test