All Projects → DeepVAC → MLab

DeepVAC / MLab

Licence: GPL-3.0 license
“云上炼丹师”中的云

Programming Languages

prolog
421 projects
Dockerfile
14818 projects
shell
77523 projects

Projects that are alternatives of or similar to MLab

konsole-snazzy
Elegant Konsole theme with bright colors
Stars: ✭ 21 (-61.11%)
Mutual labels:  kde, konsole
kali-desktop
🐳 Kali Linux desktop running in Docker on any operating system, in your web browser or a VNC client, with Kali top10 tools pre-installed.
Stars: ✭ 72 (+33.33%)
Mutual labels:  kde, vnc
plexydesk
Lightweight Desktop Manager for Gnu/Linux and FreeBSD - Ideal for Low resource computers
Stars: ✭ 33 (-38.89%)
Mutual labels:  kde
mkl-service
Python hooks for Intel(R) Math Kernel Library runtime control settings.
Stars: ✭ 45 (-16.67%)
Mutual labels:  mkl
distributed-jmeter-docker
No description or website provided.
Stars: ✭ 56 (+3.7%)
Mutual labels:  k8s
CKA-Exercises
A set of curated exercises to help prepare you for the Certified Kubernetes Administrator Exam by the Cloud Native Computing Foundation
Stars: ✭ 51 (-5.56%)
Mutual labels:  k8s
pidswallow
A swallower script using process hierarchy.
Stars: ✭ 40 (-25.93%)
Mutual labels:  kde
ubuntu-vnc-xfce-g3
Headless Ubuntu/Xfce containers with VNC/noVNC (Generation 3)
Stars: ✭ 83 (+53.7%)
Mutual labels:  vnc
kube-lineage
A CLI tool to display all dependencies or dependents of an object in a Kubernetes cluster.
Stars: ✭ 238 (+340.74%)
Mutual labels:  k8s
balanced
BalanceD is a Layer-4 Linux Virtual Server (LVS) based load balancing platform for Kubernetes.
Stars: ✭ 34 (-37.04%)
Mutual labels:  k8s
codis-operator
Codis Operator creates and manages codis clusters(proxy based Redis cluster solution) running in kubernetes.(WIP)
Stars: ✭ 21 (-61.11%)
Mutual labels:  k8s
golang-health-checker
A simple package to allow you to track your application healthy
Stars: ✭ 12 (-77.78%)
Mutual labels:  k8s
aws-cloud-map-mcs-controller-for-k8s
K8s controller implementing Multi-Cluster Services API based on AWS Cloud Map.
Stars: ✭ 61 (+12.96%)
Mutual labels:  k8s
evolution
Evolution process of The Falco Project
Stars: ✭ 37 (-31.48%)
Mutual labels:  k8s
ProductsStoreOnKubernetes
Demoing deployment of Docker containers into Kubernetes for both minikube and Azure AKS.
Stars: ✭ 90 (+66.67%)
Mutual labels:  k8s
terraform-provider-carvel
Carvel Terraform provider with resources for ytt and kapp to template and deploy to Kubernetes
Stars: ✭ 40 (-25.93%)
Mutual labels:  k8s
k8s0
Another minimal kubernetes with ansible
Stars: ✭ 23 (-57.41%)
Mutual labels:  k8s
eks-anywhere
Run Amazon EKS on your own infrastructure 🚀
Stars: ✭ 1,633 (+2924.07%)
Mutual labels:  k8s
minions
Minions for minikube - a demo of kubernetes features
Stars: ✭ 16 (-70.37%)
Mutual labels:  k8s
scheduled-scaler
No description or website provided.
Stars: ✭ 68 (+25.93%)
Mutual labels:  k8s

MLab

“云上炼丹师”中的云。

MLab构成

MLab是为云上炼丹师服务的云基础设施。由两个部分组成:

  • MLab HomePod,迄今为止最先进的容器化PyTorch训练环境。
  • MLab RookPod,迄今为止最先进的成本10万人民币以下存储解决方案。

以上两个MLab组件均为独立的产品,可以单独使用docker进行部署,也可以使用k8s进行部署。

MLab HomePod

迄今为止最先进的容器化PyTorch训练环境。

支持如下的软硬件平台:

MLab HomePod以Docker image形式(遵循OCI规范的image)封装,是我们的深度学习训练环境。目前最新版本为2.0,分为标准版和pro版本。规格如下:

参数项 MLab HomePod 2.0 MLab HomePod 2.0 pro
镜像 gemfield/homepod:2.0 gemfield/homepod:2.0-pro
OS Ubuntu 20.04 Ubuntu 20.04
PyTorch 1.9.0 1.9.0
PyTorch CUDA运行时 11.1 11.1
PyTorch CUDNN运行时 8.0.5 8.0.5
torchvision 0.10.0 0.10.0
torchaudio 0.9.0a0+33b2469 0.9.0a0+33b2469
torchtext 0.10.0 0.10.0
Conda 4.10.1 4.10.1
Python 3.8.8 3.8.8
numpy 1.20.2 1.20.2
cv2 4.5.2 4.5.2
onnx 1.8.1 1.8.1
g++ 9.3.0 9.3.0
cmake 3.16.3 3.16.3
KDE Plasma 5.22.1 5.22.1
KDE Framework 5.83.0 5.83.0
时区 中国 中国
protobuf-dev 3.6.1.3 3.6.1.3
protobuf python包 3.17.3 3.17.3
pybind11-dev 2.4.3 2.4.3
xrdp 0.9.12 0.9.12
tigervnc 1.10.1 1.10.1
VS CODE IDE 1.57.1 1.57.1
Firefox 89.0.1 89.0.1
中文输入法 IBus sunpinyin 2.0.3 IBus sunpinyin 2.0.3
coremltools 4.1 4.1
NCNN转换工具 20210525 20210525
TNN转换工具 0.3.0 0.3.0
MNN转换工具 1.2.0 1.2.0
tensorrt(转换工具) 8.0.0.3
libboost-dev 1.71.0
CUDA开发库 11.2.2
CUDNN开发库 8.1.1
MKL静态库 2020.4-912
pycuda包 2020.1
gemfield版pytorch 1.9.0
opencv4deepvac 4.4.0
libtorch静态库 1.9.0
deepvac项目 /opt/gemfield/deepvac
libdeepvac项目 /opt/gemfield/libdeepvac

除了这些核心软件,MLab HomePod还有如下鲜明特色:

  • 无缝使用DeepVAC规范;
  • 无缝构建和测试libdeepvac;
  • 包含有kdiff3、kompare、kdenlive、Dolphin、Kate、Gwenview、Konsole等诸多工具。

另外,标准版和pro版内容完全一致,除了pro版本增加了如下内容:

  • tensorrt python包,可以用来将PyTorch模型转换为TensorRT模型;
  • libboost-dev,用于C++开发者;
  • CUDA开发库,用于基于cuda的开发,或者pytorch的源码编译;
  • MKL静态库,用于基于mkl的开发,或者libtorch的静态编译;
  • pycuda python包,用于运行TensorRT模型;
  • gemfield版pytorch,基于master分支构建的pytorch python包,设置export PYTHONPATH=/opt/gemfield环境变量后来使用(从而覆盖掉标准路径下的标准版pytorch);
  • opencv4deepvac,opencv 4.4的静态库,为libdeepvac项目而生。路径为/opt/gemfield/opencv4deepvac
  • libtorch静态库,LibTorch静态库,为libdeepvac项目而生。路径为/opt/gemfield/libtorch
  • deepvac项目,https://github.com/DeepVAC/deepvac 项目在本地的克隆;
  • libdeepvac项目,https://github.com/DeepVAC/libdeepvac 项目在本地的克隆。

为了支持上述功能,pro版本的镜像足足增加了10个GB。也正是因为此,homepod从2.0版本开始拆分成了标准版和pro版。

1. 部署

MLab HomePod有三种部署方式:

  1. 纯粹的Docker命令行方式,部署且运行后只能在命令行里工作。
#有cuda设备
docker run --gpus all -it --entrypoint=/bin/bash gemfield/homepod:2.0
#没有cuda设备
docker run -it --entrypoint=/bin/bash gemfield/homepod:2.0
  1. 图形化的Docker部署方式,部署后可以在vnc客户端、rdp客户端、浏览器中访问图形界面。
#有cuda设备
docker run --gpus all -it -eGEMFIELD_MODE=VNCRDP -p 3389:3389 -p 7030:7030 -p 5900:5900 -p 20022:22 gemfield/homepod:2.0
#没有cuda设备
docker run -it -eGEMFIELD_MODE=VNCRDP -p 3389:3389 -p 7030:7030 -p 5900:5900 -p 20022:22 gemfield/homepod:2.0

参数中的端口号用途:

端口号 协议 用途
3389 rdp 用于rdp客户端,Windows远程桌面连接客户端
5900 vnc 用于vnc客户端
7030 http 用于浏览器
20022 ssh 用于ssh客户端、sftp客户端、KDE Dolphin、vscode remote ssh等

注意,当使用vscode remote ssh功能时,首先在vscode上新建ssh target,然后在"Enter SSH Connection Command"输入框中输入:

ssh -p 20022 gemfield@<your_host_running_mlab_homepod>

密码输入:deepvac

  1. k8s集群部署方式(需要k8s集群运维经验,适合团队的协作管理)。请访问基于k8s部署HomePod以获得更多部署信息。

2. 登录

三种部署方式中的第一种无需赘述,使用docker exec -it登录即可。后两种部署成功后使用图形界面进行登录和使用。支持如下使用方式:

3. 账户信息

MLab HomePod默认提供了如下账户:

  • 用户:gemfield
  • 密码:deepvac
  • HOME:/home/gemfield

如果要改变该默认行为,可以在docker命令行上(或者k8s yaml中)注入以下环境变量:

  • DEEPVAC_USER=<my_name>
  • DEEPVAC_PASSWORD=<my_password>
  • HOME=<my_home_path>

4. 账户安全

为了安全,用户在初始登录MLab HomePod后,最好使用passwd命令来修改账户密码。并在日常使用中,做到离开电脑5分钟以上手工锁定屏幕(KDE -> Leave -> Lock(Lock screen))。

MLab RookPod

迄今为止最先进的成本10万人民币以下存储解决方案。 (待补充)

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].