Facebook recently released “2018 Annual Open Source Review”, summarize and demonstrate some key data.
In 2018, Facebook opened a total of 153 new projects, with more than 94,000 new commits and 257,000 new followers.
By removing some of the archived projects, by the end of 2018, Facebook had 474 active open source projects with a total commit of 348,000 and a total follower of 1.03 million.
PyTorch, a Python-based open source deep learning platform, officially released a stable version of 1. 0 in 2018. PyTorch is currently the second fastest growing open source project on GitHub. Around the PyTorch ecosystem, Facebook opened a pair of kernel libraries —— QNNPACK and FBGEMM in 2018, which made it easier for mobile devices and servers to run the latest AI models; and PyText, a framework for accelerating NLP development.
Facebook also open sourced Horizon, an end-to-end platform built on PyTorch 1.0 and the first platform to use Application Reinforcement Learning (RL) to optimize systems in large-scale production environments. It also extends ONNX to support more AI tools. The Glow compiler is also moving towards the goal of building a machine-oriented hardware ecosystem with the support of more partners.
Facebook AI Research (FAIR) opened its object detection framework, Detectron, in 2018, and the computer vision model Mask R-CNN2Go optimized for embedded and mobile devices. Through the Open Compute Project (OCP), they also open sourced two AI-based server design specifications —— Big Sur and Big Basin. In addition, projects such as TensorComprehensions, DensePose, Translate, and TorchCraftAI are also open sourced as part of an open source framework around artificial intelligence.
In addition to working on machine learning, Facebook has had some good results in development tools, mobile, networking, data infrastructure, virtual reality and other areas in 2018. They open sourced Flipper, an extensible debugging tool for iOS and Android. For Python developers, Facebook has released the type checker Pyre and the code refactoring tool Bowler. On the network side, the scalable network load balancer Katran and the C++ 14 implementation of the TLS 1.3 standard library Fizz were released. Also released are LogDevice, a distributed data storage system designed for logs, and an XAR system for self-contained executables.
Open source Docusaurus in December 2017, a tool of choice for building websites and documents for open source projects, grew rapidly in 2018. There are currently more than 55 projects using Docusaurus, including React 360, Profilo (Performance Tracking Library), and Spectrum (Cross-Platform Image Transcoding Library).
In November, Facebook and the founders of GraphQL, the Linux Foundation, and key players in the community formed the GraphQL Foundation and managed it by the Linux Foundation.
At the end of the article, Facebook said that it will continue to invest in open source in 2019, bringing more innovative technologies to the community.