After the release of v1.8.1 , OpenPAI has entered stable mode with no major feature release planned. In order to save maintenance efforts, we changed the repo to read only mode. For collaboration, please contact repo admin directly.
With the release of v1.0, OpenPAI is switching to a more robust, more powerful and lightweight architecture. OpenPAI is also becoming more and more modular so that the platform can be easily customized and expanded to suit new needs. OpenPAI also provides many AI user-friendly features, making it easier for end users and administrators to complete daily AI tasks.
|
- When to consider OpenPAI
- Why choose OpenPAI
- Get started
- Standalone Components
- Reference
- Related Projects
- Get involved
- How to contribute
- When your organization needs to share powerful AI computing resources (GPU/FPGA farm, etc.) among teams.
- When your organization needs to share and reuse common AI assets like Model, Data, Environment, etc.
- When your organization needs an easy IT ops platform for AI.
- When you want to run a complete training pipeline in one place.
The platform incorporates the mature design that has a proven track record in Microsoft's large-scale production environment.
OpenPAI is a full stack solution. OpenPAI not only supports on-premises, hybrid, or public Cloud deployment but also supports single-box deployment for trial users.
Pre-built docker for popular AI frameworks. Easy to include heterogeneous hardware. Support Distributed training, such as distributed TensorFlow.
OpenPAI is a most complete solution for deep learning, support virtual cluster, compatible with Kubernetes eco-system, complete training pipeline at one cluster etc. OpenPAI is architected in a modular way: different module can be plugged in as appropriate. Here is the architecture of OpenPAI, highlighting technical innovations of the platform.
OpenPAI manages computing resources and is optimized for deep learning. Through docker technology, the computing hardware are decoupled with software, so that it's easy to run distributed jobs, switch with different deep learning frameworks, or run other kinds of jobs on consistent environments.
As OpenPAI is a platform, there are typically two different roles:
- Cluster users are the consumers of the cluster's computing resources. According to the deployment scenarios, cluster users could be researchers of Machine Learning and Deep Learning, data scientists, lab teachers, students and so on.
- Cluster administrators are the owners and maintainers of computing resources. The administrators are responsible for the deployment and availability of the cluster.
OpenPAI provides end-to-end manuals for both cluster users and administrators.
The admin manual is a comprehensive guide for cluster administrators, it covers (but not limited to) the following contents:
-
Installation and upgrade. The installation is based on Kubespray, and here is the system requirements. OpenPAI provides an installation guide to facilitate the installation.
If you are considering upgrade from older version to the latest v1.0.0, please refer to the table below for a brief comparison between
v0.14.0and thev1.0.0. More detail about the upgrade considerations can be found upgrade guide.v0.14.0v1.0.0Architecture Kubernetes + Hadoop YARN Kubernetes Scheduler YARN Scheduler HiveD / K8S default Job Orchestrating YARN Framework Launcher Framework Controller RESTful API v1 + v2 pure v2 Storage Team-wise storage plugin PV/PVC storage sharing Marketplace Marketplace v2 openpaimarketplace SDK Python JavaScript / TypeScript If there is any question during deployment, please check installation FAQs and troubleshooting first. If it is not covered yet, refer to here to ask question or submit an issue.
-
Basic cluster management. Through the Web-portal and a command-line tool
paictl, administrators could complete cluster managements, such as adding (or removing) nodes, monitoring nodes and services, and storages setup and permission control. -
Users and groups management. Administrators could manage the users and groups easily.
-
Alerts management. Administrators could customize alerts rules and actions.
-
Customization. Administrators could customize the cluster by plugins. Administrators could also upgrade (or downgrade) a single component (e.g. rest servers) to address customized application demands.
The user manual is a guidance for cluster users, who could train and serve deep learning (and other) tasks on OpenPAI.
-
Job submission and monitoring. The quick start tutorial is a good start for learning how to train models on OpenPAI. And more examples and supports to multiple mainstream frameworks (out-of-the-box docker images) are in here. OpenPAI also provides supports for good debuggability and advanced job functionalities.
-
Data managements. Users could use cluster provisioned storages and custom storages in their jobs. The cluster provisioned storages are well integrated and easy to configure in a job (refer to here).
-
Collaboration and sharing. OpenPAI provides facilities for collaboration in teams and organizations. The cluster provisioned storages are organized by teams (groups). And users could easily share their works (e.g. jobs) in the marketplace, where others could discover and reproduce (clone) by one-click.
Besides the webportal, OpenPAI provides VS Code extension and command line tool (preview). The VS Code extension is a friendly, GUI based client tool of OpenPAI, and it's highly recommended. It's an extension of Visual Studio Code. It can submit job, simulate jobs locally, manage multiple OpenPAI environments, and so on.
With the v1.0.0 release, OpenPAI starts using a more modularized component design and re-organize the code structure to 1 main repo together with 7 standalone key component repos. pai is the main repo, and the 7 component repos are:
- hivedscheduler is a Kubernetes Scheduler Extender for Multi-Tenant GPU clusters, which provides various advantages over standard k8s scheduler.
- frameworkcontroller is built to orchestrate all kinds of applications on Kubernetes by a single controller.
- openpai-protocol is the specification of OpenPAI job protocol.
- openpai-runtime provides runtime support which is necessary for the OpenPAI protocol.
- openpaisdk is a JavaScript SDK designed to facilitate the developers of OpenPAI to offer more user-friendly experience.
- openpaimarketplace is a service which stores examples and job templates. Users can use it from webportal plugin to share their jobs or run-and-learn others' sharing job.
- openpaivscode is a VSCode extension, which makes users connect OpenPAI clusters, submit AI jobs, simulate jobs locally and manage files in VSCode easily.
- PyTorch CIFAR-10 and TensorFlow CIFAR-10 job examples
- RESTful API
- Design documents could be found here if you are curious.
Targeting at openness and advancing state-of-art technology, Microsoft Research (MSR) and Microsoft Software Technology Center Asia (STCA) had also released few other open source projects.
- NNI : An open source AutoML toolkit for neural architecture search and hyper-parameter tuning. We encourage researchers and students leverage these projects to accelerate the AI development and research.
- MMdnn : A comprehensive, cross-framework solution to convert, visualize and diagnose deep neural network models. The "MM" in MMdnn stands for model management and "dnn" is an acronym for deep neural network.
- NeuronBlocks : An NLP deep learning modeling toolkit that helps engineers to build DNN models like playing Lego. The main goal of this toolkit is to minimize developing cost for NLP deep neural network model building, including both training and inference stages.
- SPTAG : Space Partition Tree And Graph (SPTAG) is an open source library for large scale vector approximate nearest neighbor search scenario.
- Stack Overflow: If you have questions about OpenPAI, please submit question at Stack Overflow under tag: openpai
- Gitter chat: You can also ask questions in Microsoft/pai conversation.
- Create an issue or feature request: If you have issue/ bug/ new feature, please submit it to GitHub.
This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.microsoft.com.
When you submit a pull request, a CLA-bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.
We are working on a set of major features improvement and refactor, anyone who is familiar with the features is encouraged to join the design review and discussion in the corresponding issue ticket.
- GPU fairness usage Issue 4266
- Folks who want to add support for other ML and DL frameworks
- Folks who want to make OpenPAI a richer AI platform (e.g. support for more ML pipelines, hyperparameter tuning)
- Folks who want to write tutorials/blog posts showing how to use OpenPAI to solve AI problems
One key purpose of OpenPAI is to support the highly diversified requirements from academia and industry. OpenPAI is completely open: it is under the MIT license. This makes OpenPAI particularly attractive to evaluate various research ideas, which include but not limited to the components.
OpenPAI operates in an open model. It is initially designed and developed by Microsoft Research (MSR) and Microsoft Software Technology Center Asia (STCA) platform team. We are glad to have Peking University, Xi'an Jiaotong University, Zhejiang University, University of Science and Technology of China and SHANGHAI INESA AI INNOVATION CENTER (SHAIIC) joined us to develop the platform jointly. Contributions from academia and industry are all highly welcome.