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
-
内存：773GB
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：
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 0
hccn_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：
source /usr/local/Ascend/ascend-toolkit/set_env.sh
source /usr/local/Ascend/nnal/atb/set_env.sh
source /usr/local/Ascend/atb-models/set_env.sh
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": "2",
"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"}
]
}
],
"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" : 32768,
"maxInputTokenLen" : 32768,
"truncation" : true,
"ModelConfig" : [
{
"modelInstanceType" : "Standard",
"modelName" : "deepseekr1",
"modelWeightPath" : "/workspace/deepseek-ai-r1-w8a8-mindie/",
"worldSize" : 8,
"cpuMemSize" : 5,
"npuMemSize" : -1,
"backendType" : "atb",
"trustRemoteCode" : false
}
]
},
"ScheduleConfig" : {
"templateType" : "Standard",
"templateName" : "Standard_LLM",
"cacheBlockSize" : 128,
"maxPrefillBatchSize" : 8,
"maxPrefillTokens" : 2048,
"prefillTimeMsPerReq" : 150,
"prefillPolicyType" : 0,
"decodeTimeMsPerReq" : 50,
"decodePolicyType" : 0,
"maxBatchSize" : 8,
"maxIterTimes" : 1024,
"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. 修复Tokenizer Bug
在启动服务之前，我们需要修复mies_tokenizer包中的一个Bug。以下是修复步骤：
-
查看
mies_tokenizer包的位置：pip show mies_tokenizer输出示例：
Name: mies_tokenizer Version: 0.0.1 Summary: ibis tokenizer Location: /usr/local/lib/python3.11/site-packages -
修改
tokenizer.py文件：使用
vim编辑tokenizer.py文件：vim /usr/local/lib/python3.11/site-packages/mies_tokenizer/tokenizer.py找到以下代码并进行修改：
-
修改
__del__方法：def __del__(self): cache_path = getattr(self, 'cache_path', None) if cache_path is None: return dir_path = file_utils.standardize_path(cache_path) file_utils.check_path_permission(dir_path) all_request = os.listdir(dir_path) -
修改
_get_cache_base_path方法：def _get_cache_base_path(child_dir_name): dir_path = os.getenv("LOCAL_CACHE_DIR", None) if dir_path is None: dir_path = os.path.expanduser("~/mindie/cache") os.makedirs(dir_path, exist_ok=True) os.chmod(dir_path, 0o750)
保存并退出编辑器。
-
10. 复制模型文件并检查MD5
将DeepSeek-R1模型文件复制到/data目录下，并确保目录名为deepseek-ai-r1-w8a8-mindie。
切记一定要检查md5
11. 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 &
最终资源消耗
[root@ebn0001 ~]# npu-smi info
+------------------------------------------------------------------------------------------------+
| npu-smi 24.1.0 Version: 24.1.0 |
+---------------------------+---------------+----------------------------------------------------+
| NPU Name | Health | Power(W) Temp(C) Hugepages-Usage(page)|
| Chip | Bus-Id | AICore(%) Memory-Usage(MB) HBM-Usage(MB) |
+===========================+===============+====================================================+
| 0 910B2 | OK | 95.0 37 0 / 0 |
| 0 | 0000:C1:00.0 | 0 0 / 0 62877/ 65536 |
+===========================+===============+====================================================+
| 1 910B2 | OK | 88.4 36 0 / 0 |
| 0 | 0000:C2:00.0 | 0 0 / 0 62879/ 65536 |
+===========================+===============+====================================================+
| 2 910B2 | OK | 89.1 34 0 / 0 |
| 0 | 0000:81:00.0 | 0 0 / 0 62876/ 65536 |
+===========================+===============+====================================================+
| 3 910B2 | OK | 87.8 36 0 / 0 |
| 0 | 0000:82:00.0 | 0 0 / 0 62878/ 65536 |
+===========================+===============+====================================================+
| 4 910B2 | OK | 99.8 42 0 / 0 |
| 0 | 0000:01:00.0 | 0 0 / 0 62876/ 65536 |
+===========================+===============+====================================================+
| 5 910B2 | OK | 90.6 40 0 / 0 |
| 0 | 0000:02:00.0 | 0 0 / 0 62878/ 65536 |
+===========================+===============+====================================================+
| 6 910B2 | OK | 97.5 40 0 / 0 |
| 0 | 0000:41:00.0 | 0 0 / 0 62877/ 65536 |
+===========================+===============+====================================================+
| 7 910B2 | OK | 92.8 40 0 / 0 |
| 0 | 0000:42:00.0 | 0 0 / 0 62878/ 65536 |
+===========================+===============+====================================================+
+---------------------------+---------------+----------------------------------------------------+
| NPU Chip | Process id | Process name | Process memory(MB) |
+===========================+===============+====================================================+
| 0 0 | 412122 | mindie_llm_back | 59539 |
+===========================+===============+====================================================+
| 1 0 | 412124 | mindie_llm_back | 59539 |
+===========================+===============+====================================================+
| 2 0 | 412126 | mindie_llm_back | 59539 |
+===========================+===============+====================================================+
| 3 0 | 412128 | mindie_llm_back | 59539 |
+===========================+===============+====================================================+
| 4 0 | 412140 | mindie_llm_back | 59539 |
+===========================+===============+====================================================+
| 5 0 | 412147 | mindie_llm_back | 59539 |
+===========================+===============+====================================================+
| 6 0 | 412154 | mindie_llm_back | 59539 |
+===========================+===============+====================================================+
| 7 0 | 412161 | mindie_llm_back | 59539 |
+===========================+===============+====================================================+
[root@ebn0001 ~]# free -m
total used free shared buff/cache available
Mem: 773683 359973 54194 457 359515 409456
Swap: 0 0 0
12. 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
}'
扫一扫
1582
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
