昇腾910B多机器部署DeepSeek-R1 实战指南

环境准备

硬件配置

• 节点信息：

  • node01: 10.197.61.1, 8 NPU
  • node02: 10.197.61.2, 8 NPU
  • node03: 10.197.61.3, 8 NPU
  • node04: 10.197.61.4, 8 NPU

• 内存：每台机器约773GB内存

操作系统

操作系统可以是openeuler或者ubuntu,安装包环境基本一致

• CULinux Enterprise Edition release 3.0 (Telegram)

部署步骤

1. 基础环境配置

在所有节点上执行以下命令，确保系统环境一致：

dnf update -y
dnf install -y bzip2 xz tar p7zip rsync gcc gcc-c++ make kernel-devel elfutils-libelf-devel net-tools docker-ce docker-ce-cli containerd.io

2. 安装昇腾910B驱动

进入驱动目录并安装驱动：

驱动包和固件下载最新,此包华为提供。
cd Ascend-hdk-910b-npu_24.1.0_linux-aarch64/
./install.sh query
./Ascend-hdk-910b-npu-firmware_7.5.0.3.220.run --full

安装完成后，重启机器：

reboot

3. 创建数据目录

在所有节点上创建数据目录：

mkdir /data

4. 配置NPU网络

在每个节点上配置NPU的网络信息，确保所有节点的IP不重复。以下以node01为例：

hccn_tool -i 0 -ip -s address "10.197.90.1.10" netmask 255.255.255.0
hccn_tool -i 0 -gateway -s gateway 10.197.90.254
hccn_tool -i 0 -tls -s enable 0hccn_tool -i 1 -ip -s address "10.197.90.1.11" netmask 255.255.255.0
hccn_tool -i 1 -gateway -s gateway 10.197.90.254
hccn_tool -i 1 -tls -s enable 0# 重复上述命令，配置所有NPU的网络信息

5. 下载并启动Docker镜像

下载DeepSeek-R1的Docker镜像：

docker pull swr.cn-southwest-2.myhuaweicloud.com/ei-mindie/mindie:2.0.T3-800I-A2-py311-openeuler24.03-lts

启动Docker容器：

docker run -itd --privileged --name=deepseek-r1 --net=host \--shm-size 500g \--device=/dev/davinci0 \--device=/dev/davinci1 \--device=/dev/davinci2 \--device=/dev/davinci3 \--device=/dev/davinci4 \--device=/dev/davinci5 \--device=/dev/davinci6 \--device=/dev/davinci7 \--device=/dev/davinci_manager \--device=/dev/hisi_hdc \--device /dev/devmm_svm \-v /usr/local/dcmi:/usr/local/dcmi \-v /usr/bin/hccn_tool:/usr/bin/hccn_tool \-v /usr/local/sbin:/usr/local/sbin \-v /usr/local/sbin/npu-smi:/usr/local/sbin/npu-smi \-v /usr/local/Ascend/driver:/usr/local/Ascend/driver \-v /usr/local/Ascend/firmware:/usr/local/Ascend/firmware \-v /etc/hccn.conf:/etc/hccn.conf \-v /var/log/npu/:/usr/slog \-v /sys/fs/cgroup:/sys/fs/cgroup:ro \-v /data:/workspace \swr.cn-southwest-2.myhuaweicloud.com/ei-mindie/mindie:2.0.T3-800I-A2-py311-openeuler24.03-lts \bash

进入容器：

docker exec -it <container_id> bash

6. 配置环境变量

在容器内配置环境变量，确保每个节点的MIES_CONTAINER_IP设置为当前节点的IP：

export ATB_LLM_HCCL_ENABLE=1
export ATB_LLM_COMM_BACKEND="hccl"
export HCCL_CONNECT_TIMEOUT=7200
export HCCL_EXEC_TIMEOUT=0
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export MIES_CONTAINER_IP=10.197.61.1  # 根据节点修改
export RANKTABLEFILE=/workspace/rank_table.json
export OMP_NUM_THREADS=1
export NPU_MEMORY_FRACTION=0.95

7. 创建集群配置文件

在/data目录下创建rank_table.json文件，配置多机多卡的集群信息：

{"version": "1.0","server_count": "4","server_list": [{"server_id": "10.197.61.1","container_ip": "10.197.61.1","device": [{"device_id": "0","device_ip": "10.197.90.10","rank_id": "0"},{"device_id": "1","device_ip": "10.197.90.11","rank_id": "1"},{"device_id": "2","device_ip": "10.197.90.12","rank_id": "2"},{"device_id": "3","device_ip": "10.197.90.13","rank_id": "3"},{"device_id": "4","device_ip": "10.197.90.14","rank_id": "4"},{"device_id": "5","device_ip": "10.197.90.15","rank_id": "5"},{"device_id": "6","device_ip": "10.197.90.16","rank_id": "6"},{"device_id": "7","device_ip": "10.197.90.17","rank_id": "7"}]},{"server_id": "10.197.61.2","container_ip": "10.197.61.2","device": [{"device_id": "0","device_ip": "10.197.90.20","rank_id": "8"},{"device_id": "1","device_ip": "10.197.90.21","rank_id": "9"},{"device_id": "2","device_ip": "10.197.90.22","rank_id": "10"},{"device_id": "3","device_ip": "10.197.90.23","rank_id": "11"},{"device_id": "4","device_ip": "10.197.90.24","rank_id": "12"},{"device_id": "5","device_ip": "10.197.90.25","rank_id": "13"},{"device_id": "6","device_ip": "10.197.90.26","rank_id": "14"},{"device_id": "7","device_ip": "10.197.90.27","rank_id": "15"}]},{"server_id": "10.197.61.3","container_ip": "10.197.61.3","device": [{"device_id": "0","device_ip": "10.197.90.30","rank_id": "16"},{"device_id": "1","device_ip": "10.197.90.31","rank_id": "17"},{"device_id": "2","device_ip": "10.197.90.32","rank_id": "18"},{"device_id": "3","device_ip": "10.197.90.33","rank_id": "19"},{"device_id": "4","device_ip": "10.197.90.34","rank_id": "20"},{"device_id": "5","device_ip": "10.197.90.35","rank_id": "21"},{"device_id": "6","device_ip": "10.197.90.36","rank_id": "22"},{"device_id": "7","device_ip": "10.197.90.37","rank_id": "23"}]},{"server_id": "10.197.61.4","container_ip": "10.197.61.4","device": [{"device_id": "0","device_ip": "10.197.90.40","rank_id": "24"},{"device_id": "1","device_ip": "10.197.90.41","rank_id": "25"},{"device_id": "2","device_ip": "10.197.90.42","rank_id": "26"},{"device_id": "3","device_ip": "10.197.90.43","rank_id": "27"},{"device_id": "4","device_ip": "10.197.90.44","rank_id": "28"},{"device_id": "5","device_ip": "10.197.90.45","rank_id": "29"},{"device_id": "6","device_ip": "10.197.90.46","rank_id": "30"},{"device_id": "7","device_ip": "10.197.90.47","rank_id": "31"}]}],"status": "completed"
}

8. 配置推理框架

在/data目录下创建config.json文件，配置推理框架的相关参数：

{"Version" : "1.0.0","LogConfig" : {"logLevel" : "Info","logFileSize" : 20,"logFileNum" : 20,"logPath" : "logs/mindie-server.log"},"ServerConfig" : {"ipAddress" : "10.197.61.1","managementIpAddress" : "10.197.61.1","port" : 1025,"managementPort" : 1026,"metricsPort" : 1027,"allowAllZeroIpListening" : false,"maxLinkNum" : 500,"httpsEnabled" : false,"fullTextEnabled" : false,"tlsCaPath" : "security/ca/","tlsCaFile" : ["ca.pem"],"tlsCert" : "security/certs/server.pem","tlsPk" : "security/keys/server.key.pem","tlsPkPwd" : "security/pass/key_pwd.txt","tlsCrlPath" : "security/certs/","tlsCrlFiles" : ["server_crl.pem"],"managementTlsCaFile" : ["management_ca.pem"],"managementTlsCert" : "security/certs/management/server.pem","managementTlsPk" : "security/keys/management/server.key.pem","managementTlsPkPwd" : "security/pass/management/key_pwd.txt","managementTlsCrlPath" : "security/management/certs/","managementTlsCrlFiles" : ["server_crl.pem"],"kmcKsfMaster" : "tools/pmt/master/ksfa","kmcKsfStandby" : "tools/pmt/standby/ksfb","inferMode" : "standard","interCommTLSEnabled" : false,"interCommPort" : 1121,"interCommTlsCaPath" : "security/grpc/ca/","interCommTlsCaFiles" : ["ca.pem"],"interCommTlsCert" : "security/grpc/certs/server.pem","interCommPk" : "security/grpc/keys/server.key.pem","interCommPkPwd" : "security/grpc/pass/key_pwd.txt","interCommTlsCrlPath" : "security/grpc/certs/","interCommTlsCrlFiles" : ["server_crl.pem"],"openAiSupport" : "vllm"},"BackendConfig" : {"backendName" : "mindieservice_llm_engine","modelInstanceNumber" : 1,"npuDeviceIds" : [[0,1,2,3,4,5,6,7]],"tokenizerProcessNumber" : 8,"multiNodesInferEnabled" : true,"multiNodesInferPort" : 1120,"interNodeTLSEnabled" : false,"interNodeTlsCaPath" : "security/grpc/ca/","interNodeTlsCaFiles" : ["ca.pem"],"interNodeTlsCert" : "security/grpc/certs/server.pem","interNodeTlsPk" : "security/grpc/keys/server.key.pem","interNodeTlsPkPwd" : "security/grpc/pass/mindie_server_key_pwd.txt","interNodeTlsCrlPath" : "security/grpc/certs/","interNodeTlsCrlFiles" : ["server_crl.pem"],"interNodeKmcKsfMaster" : "tools/pmt/master/ksfa","interNodeKmcKsfStandby" : "tools/pmt/standby/ksfb","ModelDeployConfig" : {"maxSeqLen" : 20000,"maxInputTokenLen" : 20000,"truncation" : true,"ModelConfig" : [{"modelInstanceType" : "Standard","modelName" : "deepseekr1","modelWeightPath" : "/workspace/DeepSeek-R1-bf16-mind/","worldSize" : 8,"cpuMemSize" : 5,"npuMemSize" : -1,"backendType" : "atb","trustRemoteCode" : false}]},"ScheduleConfig" : {"templateType" : "Standard","templateName" : "Standard_LLM","cacheBlockSize" : 128,"maxPrefillBatchSize" : 8,"maxPrefillTokens" : 2024,"prefillTimeMsPerReq" : 150,"prefillPolicyType" : 0,"decodeTimeMsPerReq" : 50,"decodePolicyType" : 0,"maxBatchSize" : 8,"maxIterTimes" : 2024,"maxPreemptCount" : 0,"supportSelectBatch" : false,"maxQueueDelayMicroseconds" : 5000}}
}

将config.json文件覆盖到MindIE服务的配置目录：

cp /workspace/config.json /usr/local/Ascend/mindie/latest/mindie-service/conf/

9. 复制模型文件

将DeepSeek-R1模型文件复制到/data目录下，并确保目录名为DeepSeek-R1-bf16-mind。

10. 启动服务

在所有节点上按照顺序启动服务：

nohup /usr/local/Ascend/mindie/latest/mindie-service/bin/mindieservice_daemon > /workspace/output.log 2>&1 &

11. 测试接口

通过以下命令测试接口是否正常工作：

curl -X POST http://10.197.61.1:1025/v1/chat/completions \
-H "Accept: application/json" \
-H "Content-Type: application/json" \
-d '{"model": "deepseekr1","messages": [{"role": "user","content": "你好"}],"max_tokens": 20,"top_p": 0.95
}'

总结

通过以上步骤，我们成功在四台搭载昇腾910B的服务器上部署了DeepSeek-R1模型，并配置了多机多卡环境。希望本文能为有类似需求的开发者提供参考。