foundationpose Ubuntu22.04复现(一)胎教级配环境

foundationpose +Ubuntu22.04复现(一)配置环境

配置:ubuntu22.04+RTX3080

cuda12.1
python3.9
pytorch 2.4.0
pytorch3d-0.7.8

首先介绍一个很好用的东西:timeshift,可以保存系统快照,在某些时刻你配环境配崩的时候可以回到过去的某一刻,重新来过,非常好用,不用再经历繁琐的重装系统过程

1.NIVIDIA–driver

attention:

如果你选择安装NVIDIA driver545,则需要更新gcc版本,22.04系统默认配置为gcc-11,需要更新为gcc-12,

这里有一个大坑:后续安装的cuda必须为12.x,11.x的只支持gcc-11,因为我一开始是NVIDIA driver545+gcc-12+cuda11.8的搭配,后面编译文件的时候出现报错说cuda11.8不支持gcc-12编译,仅支持gcc-11,所以只能降级NVIDIA driver为更低版本或者升级cuda。

此时我面临着一个难题:不知道出何原因,我的系统一旦安装了NVIDIA driver535,就会出现莫名其妙的掉网卡,图形界面分辨率降低的情况(不过我还没试过NVIDIA driver530)

所以我选择了升级cuda为12.1,若要重装cuda,可以参考这篇博客https://blog.youkuaiyun.com/diandianing/article/details/135667829

更新gcc版本

sudo apt-get update

sudo apt-get install gcc-12 g++-12

将gcc-12设置为默认配置

sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-12 100 --slave /usr/bin/g++ g++ /usr/bin/g++-12

查看版本:gcc --version

前置准备(选做

lsb_release -a(系统架构)

lspci | grep -i vga(所有显卡)

lspci -v -s 00:02.0 (输入显卡序号,获得某块显卡的详细信息)

查看显卡类型:https://admin.pci-ids.ucw.cz/mods/PC/10de/2204

方式一:

下拉菜单查看“软件和更新–>附加驱动“中查看此电脑支持的显卡驱动
在这里插入图片描述

方式二:

在官网下载类似于NVIDIA-Linux-x86_64-545.29.06.run的文件,注意显卡型号和所对应的驱动版本(可有多个)

https://www.nvidia.com/en-us/drivers/unix/

给文件赋权限:

chmod 777 NVIDIA-Linux-x86_64-545.29.06.run

运行:

sudo ./NVIDIA-Linux-x86_64-545.29.06.run

*配置过程中会有选项让你生成禁用 nouveau的文件,选择

### Deepseek-v3 on Ubuntu Installation and Usage For installing and using Deepseek-v3 on an Ubuntu system, several preparatory steps are necessary to ensure that the environment is correctly set up. The following details cover these aspects comprehensively. #### Preparing the Environment To prepare the environment for Deepseek-v3, it's essential first to have a properly configured development setup which includes updating package lists and ensuring all dependencies required by Deepseek-v3 or similar projects are installed: Updating packages ensures you get the latest versions of software available in your distribution’s repositories[^3]: ```bash sudo apt-get update ``` Installing build tools such as `gcc`, `make` among others can be done with this command[^2]: ```bash sudo apt-get install gcc g++ make cmake libpcre2-dev libpcre3-dev libssl-dev zlib1g-dev perl m4 libtool automake autoconf vim-common unzip libcap2-bin ``` These commands will provide a solid foundation needed not only potentially for building components from source but also supporting various operations within Deepseek-v3. #### Installing Python Dependencies Python-based applications like Deepseek-v3 often rely heavily upon specific libraries defined in a file named `requirements.txt`. To efficiently manage installations while reducing latency due to network requests during dependency resolution, one might opt to use a mirror closer geographically located compared to default servers provided globally. An example would involve utilizing Huazhong University of Science and Technology (HUST)'s PyPI mirror server when executing pip installs via Python 3[^1]: ```bash python3 -m pip install -r requirements.txt -i https://mirrors.hust.edu.cn/pypi/web/simple ``` This approach helps speed up downloads significantly especially under conditions where international bandwidth may pose limitations affecting overall efficiency negatively otherwise experienced without optimization measures taken into account beforehand. #### Specific Steps for Deepseek-v3 While detailed instructions specifically tailored towards setting up Deepseek-v3 aren't directly covered here, based on common practices observed across many machine learning frameworks deployed over Linux distributions including those mentioned previously about preparing environments adequately before proceeding further – users should refer closely any official documentation associated particularly with Deepseek-v3 releases targeting Ubuntu platforms. This typically involves cloning repository sources related thereto followed by configuring according to guidelines outlined therein concerning compilation parameters alongside other settings pertinent specifically toward achieving successful deployments thereof. --related questions-- 1. What additional configurations must be considered for optimizing performance after completing basic installation procedures? 2. How does changing mirrors impact security considerations regarding trustworthiness of downloaded packages? 3. Are there alternative methods recommended besides using pip for managing project-specific virtual environments containing isolated sets of dependencies? 4. Can certain parts of preparation processes described above be automated through scripts or integrated toolchains designed explicitly around streamlining deployment workflows involving multiple interdependent services?
评论 6
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值