TechPrufer
Get started
Meta
Meta logoMeta · Feature trail

Open Source Strategy and Contributions

32 posts · Jun 2008 – Dec 2025

What this trail covers

Meta's open source strategy continues to evolve with the release of Ax 1.0, an adaptive experimentation platform leveraging Bayesian optimization. Ax is used across Meta for hyperparameter optimization, GenAI data mixture discovery, production infrastructure tuning, and AR/VR hardware design. The platform empowers users to configure and run dynamic experiments, providing analysis for understanding system behavior and Pareto frontiers. Ax integrates with BoTorch for Bayesian optimization components. This post describes the application of Bayesian optimization for tuning online systems using A/B tests, which is a core component of experimentation platforms.

All Meta trails
  1. Python Typing Survey 2025: Code Quality and Flexibility As Top Reasons for Typing AdoptionThis post details the findings of the 2025 Typed Python Survey, conducted by contributors from JetBrains, Meta, and the broader Python typing community. It highlights key trends in Python typing adoption, developer attitudes, challenges, and requested features. The survey indicates strong adoption of type hints, with 86% of respondents using them 'always' or 'often'. Key reasons for adoption include improved readability, tooling support, and bug prevention. Challenges identified include third-party library support, complexity of advanced features, and tooling fragmentation. Requested features mirror those found in languages like TypeScript, with a desire for runtime enforcement and better generics. The post also touches on tooling trends, with Mypy remaining dominant but Rust-based checkers gaining traction, and discusses how developers learn and get help, with official documentation, blog posts, and community platforms being primary sources, alongside growing use of LLMs.
  2. Efficient Optimization With Ax, an Open Platform for Adaptive ExperimentationRelease of Ax 1.0, an open-source adaptive experimentation platform utilizing Bayesian optimization (leveraging BoTorch) for efficient hyperparameter tuning and system optimization. Introduces concepts of exploration vs. exploitation, surrogate models (Gaussian processes), and acquisition functions (expected improvement). Details applications in ML, GenAI, infrastructure tuning, and hardware design. Provides analysis tools for understanding system behavior and Pareto frontiers. Encourages community contributions.
  3. Open Source Is Good for the EnvironmentThis post highlights Meta's expansion of its open-source strategy to include open hardware and its application in environmental sustainability. It details Meta's participation in the 2025 Open Compute Project (OCP) Summit, the introduction of a new AI methodology for understanding Scope 3 emissions, and the use of AI in developing sustainable data center construction materials, all contributing to Meta's net-zero emissions goal.
  4. Meta joins Kotlin FoundationMeta has joined the Kotlin Foundation as a gold member, signifying a commitment to the Kotlin language and the Android development ecosystem. This post details Meta's ongoing efforts to migrate its extensive Android codebase from Java to Kotlin, including the development of the internal tool Kotlinator for automated conversion. It also highlights contributions to the Kotlin and Android build toolchain within Buck2, aiming to improve build speeds and scalability. Meta will also contribute to the Kotlin Foundation's grants program to support open-source library authors and developers.
  5. Open-sourcing Pyrefly: A faster Python type checker written in RustThis post announces the open-sourcing of Pyrefly, a new Python type checker written in Rust. It highlights the evolution of this tool from an internal project for Instagram's typed Python codebase to a community-available resource, emphasizing its role in improving Python development tooling and catching errors before runtime. It also mentions the accompanying Meta Tech Podcast episode discussing its development.
  7. Introducing Pyrefly: A new type checker and IDE experience for PythonThis post introduces Pyrefly, a new open-source Python type checker and IDE extension developed in Rust. It highlights Meta's ongoing commitment to improving the Python ecosystem and developer experience through open-source contributions. Pyrefly builds upon previous efforts like Pyre, focusing on enhanced performance, IDE integration, and type inference. The post emphasizes collaboration with the Python community and outlines plans for future development and contributions to language enhancements.
  8. Enhancing the Python ecosystem with type checking and free threadingThis post details Meta's contributions to the Python ecosystem, specifically focusing on improving type annotation coverage for pandas-stubs (increasing it from 36% to over 50%) and enhancing compatibility with free-threaded Python (FTP) for popular scientific computing and language binding packages. It also highlights the development of tooling to catalog missing annotations and measure type coverage for independently distributed stubs.
  9. Threads has entered the fediverseThis post details Meta's entry into the fediverse with Threads, marking a significant expansion of its open-source strategy. It explains the adoption of the ActivityPub protocol for interoperability, the technical challenges and solutions involved in federating social content, and the phased approach to integrating Threads with decentralized social networks. This represents a new dimension of Meta's open-source contributions, moving beyond tools and infrastructure to embrace open social networking protocols.
  10. Meta loves PythonThis post details Meta engineers' contributions to Python 3.12, including the development of custom JITs (Cinder), Immortal Objects, type system improvements, and faster comprehensions. It highlights the technical reasons behind these developments and Meta's engagement with the Python community.
  11. DotSlash: Simplified executable deploymentThis post introduces and open-sources DotSlash, a tool designed to simplify the deployment of large executables within source control by replacing them with lightweight scripts. It addresses challenges related to monorepo size, clone operations, and the management of numerous first-party and third-party command-line tools. The post details DotSlash's functionality, its internal usage at Meta (hundreds of millions of executions daily), and its cross-platform capabilities, along with a complementary GitHub Action for external use. It also explains the generation of DotSlash files through automated build pipelines.
  12. Meta contributes new features to Python 3.12Meta contributed several features and optimizations to Python 3.12, including Immortal Objects (PEP 683) to reduce memory usage and enable better parallelism, the @typing.override decorator (PEP 698) for improved type safety during refactoring, and performance optimizations such as inlining comprehensions (PEP 709), eager asyncio tasks, faster super() calls, and optimizations for hasattr and unittest.mock.Mock. They also added new benchmarks to the Python Performance Benchmark suite to better represent Meta's workloads and landed core runtime hooks for their Cinder JIT compiler and Static Python, including APIs for function entrypoints, dictionary/type/function/code object watchers, and extensibility in the code generator.
  13. Open sourcing Winterfell: A STARK prover and verifierThis post announces the open-sourcing of Winterfell, a Rust implementation of a STARK prover and verifier, to Crates.io. Winterfell is designed to make zero-knowledge proofs (ZKPs) more accessible to average developers by abstracting away cryptographic complexity. It leverages STARKs for their transparency, minimal cryptographic assumptions, and potential for quantum resistance. The release aims to support applications in blockchain privacy and scalability, as well as other security and privacy use cases.
  14. Open-sourcing Thrift for HaskellOpen-sourced hsthrift, a Haskell Thrift compiler and libraries, enabling cross-service communication with other fbthrift languages. Introduced advanced features like required symbols pruning, exactprint for AST transformations with formatting preservation, typechecker plugins for backend extensibility, and splicing for typeclass instances.
  15. Infer powering Microsoft’s Infer#, a new static analyzer for C#This post details the creation of Infer#, a new language front end for Meta's Infer static analysis platform, enabling it to analyze C# code. This collaboration with Microsoft extends the reach of Infer to the .NET platform, allowing it to detect deep inter-procedural bugs in C# code, similar to its capabilities in Java and C++. It also introduces a language-independent JSON layer for SIL, paving the way for future Infer front ends for other languages.
  16. CG/SQL: Easy, accurate code generation for SQLiteThis post introduces CG/SQL, an open-source code generation system for SQLite. It allows developers to write stored procedures in a T-SQL variant and compile them into C code, significantly improving the accuracy and maintainability of data access layers. The system includes features for schema management, testing, and query plan analysis, making complex SQLite operations more robust and easier to manage.
  17. Data Transfer Project: Enabling portability of photos and videos between servicesThis post announces Meta's participation in and the release of a tool for the Data Transfer Project, a collaborative open-source effort with other major tech companies to enable service-to-service data portability. It details the project's mission, architecture (shared data models, adapters, task management framework), and deployment at Facebook, emphasizing security, reliability, and custom extensions leveraging internal infrastructure like TAO and Scribe. The post highlights the release of a tool for transferring photos and videos from Facebook to Google Photos, with plans for future expansion.
  18. Simon Marlow, Simon Peyton Jones, and Satnam Singh win Most Influential ICFP Paper AwardThis post highlights the significant impact of an influential ICFP paper on "Runtime Support for Multicore Haskell," co-authored by a Facebook engineer. It demonstrates how foundational research, leading to improvements in a programming language (Haskell) and its compiler (GHC), can be leveraged for large-scale, critical systems like Facebook's abuse protection (Sigma). This showcases how Meta contributes to and benefits from advancements in programming language technology, which can then be applied to solve complex engineering challenges and drive efficiency in production systems, aligning with the broader theme of open-source contributions and the adoption of advanced technologies.
  19. Hydra: A framework that simplifies development of complex applicationsThis post introduces Hydra, a lightweight Python framework for simplifying the development of complex applications by enabling composable and overridable configurations. It addresses common challenges like boilerplate code, configuration management, and command-line flag definition. The post highlights its use in prototyping research projects at Facebook and its open-sourcing to benefit the wider industry, aligning with Meta's broader open-source strategy.
  20. Facebook’s open source browser contributionsThis post details Meta's contributions to the web platform and browser technologies, specifically focusing on projects like isInputPending(), the JavaScript Self-Profiling API, and BinAST. These contributions aim to improve web application performance by addressing limitations in JavaScript code size, parsing overhead, and introspection APIs, demonstrating Meta's commitment to open-source development beyond its core AI/ML and infrastructure tools.
  21. Open source at F8 2019This post announces the open-sourcing of several tools at F8 2019: Ax (adaptive experimentation platform), BoTorch (Bayesian optimization research library), idb (iOS simulator/device automation CLI), Memscout (jemalloc memory analysis tool), and Mvfst (QUIC transport protocol implementation).
  22. Open-sourcing F14 for faster, more memory-efficient hash tablesThis post introduces F14, a new 14-way probing hash table implementation within the Folly open-source library. F14 aims to simplify hash table selection by offering a default, highly optimized solution that balances CPU and memory performance. It leverages vector instructions for faster collision resolution and introduces a novel reference-counted tombstone-like mechanism for efficient handling of insertions and deletions, improving both performance and memory utilization across various use cases.
  23. Open-sourcing Habana back end for GlowThis post announces the open-sourcing of an experimental back end for the Glow compiler, specifically designed to target Habana's hardware accelerators. This marks a significant step in Glow's evolution by being the first back end customized for a specific vendor's hardware, thereby strengthening the ecosystem for ML hardware acceleration and enabling partners to optimize silicon products for machine learning workloads.
  24. Open-sourcing Python Test Runner for multiple tests in parallelIntroduced `ptr` (Python Test Runner), an open-source tool that crawls repositories to find Python projects with unit tests, runs them in parallel, and supports code coverage checks, formatting, linting, and static type analysis. This enhances the efficiency of testing Python codebases.
  25. Team behind Infer wins POPL 2019 Most Influential Paper AwardThis post details the open-sourcing of Infer, a static program analyzer, and its associated declarative language AL. It highlights the award-winning research behind Infer, its impact on detecting bugs at scale in Meta's mobile code, and its adoption by other companies, further solidifying Meta's commitment to open-sourcing impactful developer tools.
  26. Open source year in reviewThis post details Meta's open source contributions in 2018, highlighting the release of 153 new projects and a total of 474 active projects. It showcases significant releases in AI/ML (PyTorch 1.0, QNNPACK, FBGEMM, PyText, Horizon, ONNX, Glow, Detectron, Mask R-CNN2Go, Big Sur/Basin, TensorComprehensions, DensePose, Translate, TorchCraftAI), development tools (Flipper, Pyre, Bowler), networking (Katran, Fizz), data infrastructure (LogDevice, XAR), and documentation tools (Docusaurus). It also mentions the formation of the GraphQL foundation and improvements to internal tooling like ShipIt.
  27. Efficient tuning of online systems using Bayesian optimizationThis post describes the application of Bayesian optimization to adaptively design rounds of A/B tests for tuning parameters in online systems, particularly machine learning systems. It highlights that Bayesian optimization allows for tuning more parameters with fewer experiments compared to grid search or manual tuning, and has been used for dozens of parameter tuning experiments across backend systems.
  28. Facebook Open Source 2016 year in reviewThis post details Meta's open source activities in 2016, highlighting the launch of 77 new projects and 60,000 commits. It emphasizes the growth and adoption of flagship projects like React and React Native, and introduces new projects such as Draft.js and create-react-app. The release of tools for Android and iOS, Zstandard compression, and the collaborative development of Yarn are also noted. The post also covers contributions to external projects like Chef, systemd, and the Linux kernel, as well as open-sourcing AI resources like fastText and the Surround360 camera system. It highlights collaborations with Microsoft, Samsung, and Spotify, and GitHub's adoption of GraphQL.
  29. What’s new in Facebook open sourceThis post announces the addition of 54 new open-source projects in the first half of 2016, highlighting Draft.js (rich text editor framework), ReDex (Android bytecode optimizer), and Memory Bundle (iOS memory profiling tools) as notable examples. It also mentions open-sourcing Torch implementations for deep neural nets, Torchnet, and Facebook CTF for security competitions, as well as Reason, a new interface to OCaml. The post emphasizes continued support for existing projects and community building around them.
  30. Facebook open sources rich text editor framework Draft.jsThis post details the open-sourcing of Draft.js, a React-based rich text editor framework. It explains the technical challenges of traditional rich text input (e.g., using <textarea>), the decision to build a custom solution integrated with React, and the architectural choices made, including the use of a controlled ContentEditable component and immutable data structures for state management and undo/redo functionality. The post highlights the benefits of this approach for developer experience, user experience, and memory efficiency.
  31. Open source in 2015: A year of growthThis post details the growth of Meta's open source program in 2015, highlighting increased adoption of projects like React, HHVM, and Presto by external companies. It notes the significant growth in community contributions, with over 2,500 external contributors and more than 50,000 commits to public projects. New projects like React Native, Relay, and GraphQL were launched and gained substantial traction. The post emphasizes the value of community contributions and outlines a commitment to continued support for open source projects and their developers in 2016.
  32. 2013: A Year of Open Source at FacebookThis post details Meta's significant open source contributions in 2013 across mobile, web, data, and infrastructure. Key new releases include Buck (Android build tool), xctool (iOS/Mac build/test tool), Rebound (Android animation library), and Origami (mobile interaction prototyping toolkit). On the web front, React, Regenerator, and Huxley were highlighted. For data, Presto (distributed SQL query engine) and RocksDB (embedded key-value store) were launched. Infrastructure saw major updates to HHVM, along with new projects like pfff, libPhenom, and folly, and contributions to Mercurial, LLVM, and GNU grep. The post also outlines Meta's renewed commitment to actively maintaining its open source projects, improving community engagement, and supporting open source education through the Facebook Open Academy and '@ Scale' events.
  33. Adapting Open Source SoftwareThis post details how Facebook adapts open source software like PHP, MySQL, and Memcache to its specific needs, and contributes technologies like Thrift and Scribe back to the community. It highlights custom backend services for NewsFeed and Search, and the tooling stack that enables engineers to develop, deploy, monitor, and maintain code in various languages.
  34. Thrift: (slightly more than) one year laterThis post announces the one-year anniversary of Thrift's open-sourcing and highlights its evolution. New features include support for C#, Perl, Objective C, Erlang, Smalltalk, OCaml, and Haskell, improved Java and Ruby mappings, two new protocols (dense encoding and JSON), and significant speed boosts in Python and PHP. External contributions from companies like Powerset, imeem, Evernote, and Amie Street are noted. Future work includes an asynchronous C++ client/server, templating the C++ API for performance gains, and a C mapping using glib. Thrift's acceptance into the Apache Incubator is also a major development, signaling its potential integration into the broader Apache ecosystem.