# corex_knife

**Repository Path**: 121786404/corex_knife

## Basic Information

- **Project Name**: corex_knife
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 3
- **Created**: 2025-03-30
- **Last Updated**: 2026-03-13

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# 1. 对服务器进行设置

    参考 driver/服务器硬件设置.md ,设置完成后运行tools/x86(arm)/ix-gpu-tuner检查
    ix-gpu-tuner --trace 

    ![image.jpg](https://gitee.com/121786404/corex_knife/raw/master/tools/assets/ix-gpu-tuner_0.jpg)

# 2. 安装驱动

```
    #安装必须的包

    #ubuntu
    apt install linux-headers-$(uname -r) gcc make

    #centos
    yum install kernel-devel gcc make
```
```
    #锁定内核版本不要升级
    (ubuntu)
    apt-mark hold linux-image-generic linux-headers-generic linux-headers-$(uname -r) linux-image$(uname -r) linux-modules$(uname -r) linux-modules-extra$(uname -r)
    (centos)
    修改/etc/yum.conf，最后面加入 exclude=kernel*
```
``` 
    #安装头文件
    wget https://gitee.com/121786404/corex_knife/raw/master/driver/partial_install_cuda_header.tar.gz
    tar xf partial_install_cuda_header.tar.gz
    cd partial_install_cuda_header
    bash install-cuda-header.sh 
```
``` 
    #卸载旧的驱动
    /usr/local/corex/bin/corex-driver-uninstaller 
```
```
    #安装新驱动
    (X86) bash corex-driver-linux64-4.2.0_x86_64_10.2.run
    (ARM) bash corex-driver-linux64-4.3.6_aarch64_10.2.run
    
    wget https://gitee.com/121786404/corex_knife/raw/master/driver/corex.conf
    cp corex.conf /etc/ld.so.conf.d/
    echo "export PATH=\$PATH:/usr/local/corex/bin" >> /etc/profile
    rm -f /usr/local/corex
    ln -s /usr/local/corex-* /usr/local/corex
    ldconfig
    执行 ixsmi 可以看到GPU信息
```
## 驱动安装常见报错

 (1) Error code:32 message: Kernel source is not ready   
 运行```rpm -q kernel-devel```和```uname -a```，对比内核版本是否一致，若不一致， 运行   
 ```
 #yum list kernel-devel --showduplicates  
 #yum remove kernel-devel-xxx (rpm -q kernel-devel打印的内核版本后缀)
 yum install kernel-devel-$(uname -r | cut -d'.' -f1-5)
 
 如果提示找不到，可以手动下载
 
 openeuler 
 cat /etc/os-release   查看openEuler版本
 https://mirror.nju.edu.cn/openeuler/openEuler-24.03-LTS/everything/aarch64/Packages/kernel-devel-*.rpm

	kylin
 https://update.cs2c.com.cn/NS/V10/V10SP3-2403/os/adv/lic/base/x86_64/Packages/kernel-devel-4.19.90-89.11.v2401.ky10.x86_64.rpm
 
 ```

# 3.安装docker-ce并拉取镜像
```
    联网安装 https://developer.aliyun.com/mirror/docker-ce
    离线安装 https://docs.docker.com/engine/install/binaries/#install-daemon-and-client-binaries-on-linux
```
```
    docker run --name="corex_4.3.0_v10" --ulimit memlock=-1:-1 -itd --privileged --cap-add=ALL --network=host --ipc=host --pid=host -v /dev:/dev -v /lib/modules:/lib/modules -v /usr/src:/usr/src -v /home/:/home -v /root:/root crpi-92uj7jb20gffz04j.cn-guangzhou.personal.cr.aliyuncs.com/iluvatar_common/vllm0.8.3-4.3.0-x86:v10 /bin/bash 

    docker run --name="corex_4.3.6_v10" --ulimit memlock=-1:-1 -itd --privileged --cap-add=ALL --network=host --ipc=host --pid=host -v /dev:/dev -v /lib/modules:/lib/modules -v /usr/src:/usr/src -v /home/:/home -v /root:/root crpi-92uj7jb20gffz04j.cn-guangzhou.personal.cr.aliyuncs.com/iluvatar_common/vllm0.8.3-4.3.6-aarch64:v10 /bin/bash

    运行 python3 -c "import torch;print(torch. cuda. is_available())"  返回true，表示pytorch可以调用到gpu了
```