Preparation of the basic environment
- hardware requirement：NVIDIA H800等支持NVLink/RDMA的GPU
- network requirement：InfiniBand 400Gb/s以上
- software dependency：CUDA 11+/NCCL/Python 3.8+
Step-by-step installation guide
- 克隆源码库：
git clone https://github.com/deepseek-ai/DeepEP.git - 安装修改版NVSHMEM（需应用补丁文件）
- 执行编译：
make生成内核文件 - Setting environment variables:
export NVSHMEM_IB_SL=0export NVSHMEM_ENABLE_ADAPTIVE_ROUTING=1(限低延迟模式)
Verify Installation
Run the test script:python tests/test_low_latency.py
预期输出应包含：
- All-to-all通信成功标志
- 各GPU间的带宽测试结果
- 时延统计信息
This answer comes from the articleDeepEP: An Open Source Tool to Optimize Communication Efficiency Specifically for MoE Models (DeepSeek Open Source Week Day 2)The
May not be reproduced without permission:AI productivity tools " 如何安装和配置DeepEP的开发环境？
English 
简体中文 
日本語 
Deutsch 
Português do Brasil