MindIE部署
部署环境
显卡:Atlas300I Duo
系统:Ubuntu 20.04 内核版本
docker: 26.1.3
驱动:Ascend-hdk-310p-npu-driver_24.1.rc2_linux-aarch64.run
固件:Ascend-hdk-310p-npu-firmware_7.3.0.1.231.run
MindIE: 1.0.0-300I-Duo-py311-openeuler24.03-lts
安装驱动固件安装
环境准备
# 安装dkms、gcc、linux-header等依赖
sudo apt update && sudo apt install -y gcc build-essential linux-headers-generic dkms
配置用户组
groupadd HwHiAiUser
useradd -g HwHiAiUser -d /home/HwHiAiUser -m HwHiAiUser -s /bin/bash
安装固件和驱动
# 1 赋可执行权限
chmod +x Ascend-hdk-310p-npu-driver_24.1.rc2_linux-aarch64.run
chomd +x Ascend-hdk-310p-npu-firmware_7.3.0.1.231.run
# 2 检查
./Ascend-hdk-310p-npu-driver_24.1.rc2_linux-aarch64.run --check
./Ascend-hdk-310p-npu-firmware_7.3.0.1.231.run --check
# 3 安装
./Ascend-hdk-310p-npu-driver_24.1.rc2_linux-aarch64.run --full
./Ascend-hdk-310p-npu-firmware_7.3.0.1.231.run --full
# 4 验证
npu-smi info
# 5 版本升级(按需)
./Ascend-hdk-所需版本-firmware.run --upgrade
./Ascend-hdk-所需版本-driver.run --upgrade
安装MindIE
登录
docker login -u cn-south-1@SY5LJAU8345CLNZH11P1 swr.cn-south-1.myhuaweicloud.com
输入密码
拉取镜像
docker pull swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:1.0.0-300I-Duo-py311-openeuler24.03-lts
运行容器
# 按需修改容器名、镜像名、所用设备--device、权重位置 -v /usr/local/deepseek:/storage/llm
docker run -it --name ds70b --ipc=host --net=host --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/devmm_svm --device=/dev/hisi_hdc -v /usr/local/Ascend/driver:/usr/local/Ascend/driver -v /usr/local/Ascend/add-ons/:/usr/local/Ascend/add-ons/ -v /usr/local/sbin/:/usr/local/sbin/ -v /usr/local/sbin/npu-smi:/usr/local/sbin/npu-smi -v /var/log/npu/conf/slog/slog.conf:/var/log/npu/conf/slog/slog.conf -v /var/log/npu/slog/:/var/log/npu/slog -v /var/log/npu/profiling/:/var/log/npu/profiling -v /var/log/npu/dump/:/var/log/npu/dump -v /etc/hccn.conf:/etc/hccn.conf -v /etc/localtime:/etc/localtime -v /usr/local/dcmi:/usr/local/dcmi -v /etc/ascend_install.info:/etc/ascend_install.info -v /etc/vnpu.cfg:/etc/vnpu.cfg -v /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info -v /usr/local/deepseek:/storage/llm -w /storage/llm cded703a9282 bash
部署DeepSeek-R1-Distill-Llama-70B
进入容器
docker exec -it 容器名/容器ID /bin/bash
加载环境
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
source /usr/local/Ascend/mindie/set_env.sh
编辑配置
# 编辑,配置
cd /usr/local/Ascend/mindie/1.0.0/mindie-service/conf
vi config.json
# 赋权
chmod 640 config.json
执行
cd /usr/local/Ascend/mindie/1.0.0/mindie-service/bin
./mindieservice_daemon
测试
curl -H "Accept: application/json" -H "Content-type: application/json" -X POST -d '{
"prompt": "1+1=?",
"max_tokens": 512,
"repetition_penalty": 1.03,
"presence_penalty": 1.2,
"frequency_penalty": 1.2,
"temperature": 0.5,
"top_k": 10,
"top_p": 0.95,
"stream": false
}' http://127.0.0.1:1025/generate
问题
1、如果遇到有数据格式的报错,修改下载权重文件夹下config.json文件,将"torch_dtype": “bfloat16"
改成"torch_dtype”: “float16”
2、如果遇到有权限的报错,那可能是模型权重路径的权限太高了,会报错。执行如下命令: chmod 640 /your_ckpt_path -R
修改权重路径的权限。
Last modified: 12 March 2025