Meta Open Source

🔎 Meta Open Source 101🔎 Do you work with survey data that doesn't quite match your target audience? Use Balance, a Python package that adjusts biased data samples (like non-representative surveys) by creating weights to match a target population. Surveys often suffer from selection bias—where the people who respond don't match the general population—but Balance fixes this by fitting weights to your sample using standard statistical methods like Raking, Post-stratification, and Propensity Score weighting (IPW). The workflow is designed to be simple and data-driven: you load your sample and target data, run sample.adjust() to fit weights automatically, and then evaluate the results using sample.covars().plot() to visualize the reduction in bias. Balance is used internally at Meta to ensure accurate insights from user sentiment surveys, empowering developers to use rigorous statistical methodologies without needing a deep background in survey statistics. Learn more at https://lnkd.in/epcqhUXc and read the deep dive here: https://lnkd.in/eN8hqYKb

⚡️OSS Project Spotlight⚡️ Folly is a library of C++ components designed with practicality and efficiency in mind. It complements the standard library with optimized utilities—such as fast synchronization primitives and memory-efficient containers—that power Meta's core infrastructure at scale. Learn more: https://lnkd.in/dt-ZbyeN

🔎 Meta Open Source 101🔎 Need the development speed of a dynamic language but the discipline of a static one? Meet HHVM (HipHop Virtual Machine) and Hack, the engine and language designed to execute PHP code with lightning-fast performance using Just-In-Time (JIT) compilation. Hack introduces "gradual typing," allowing you to mix dynamic and static typing seamlessly. This means you can write code quickly like in PHP, but use Hack's type checker (hh_client) to catch errors instantly before you even run the program—preventing crashes in production. Running a script is simple. Instead of the standard PHP engine, you use the HHVM CLI. To run a file, simply use hhvm script.hack, or to start a web server on port 8080, run hhvm -m server -p 8080. HHVM powers the massive web tier at Meta, handling billions of requests, and evolved to support the specific needs of modern web development at extreme scale. Learn more at https://hhvm.com/ and read the engineering deep dive here: https://lnkd.in/ecnQy9kn

⚡️OSS Project Spotlight⚡️ Lexical is an extensible text editor framework designed for reliability, accessibility, and performance. It provides a powerful, framework-agnostic engine for building sophisticated text editing experiences on both the web and iOS. Learn more: https://lexical.dev/

New Meta Open Source Content 🔎 Four type-narrowing patterns that make Python type checking more intuitive. 🐍✨ From tuple length narrowing to hasattr guards, see how Pyrefly helps reduce the need for explicit casts in your code. Learn more here: https://lnkd.in/eAny9r5R

🔎 Meta Open Source 101🔎 Want to catch null pointer exceptions and resource leaks before they crash your app? Use Infer, a static analyzer that detects complex bugs in Java, C++, and Objective-C. Unlike simple linters, Infer performs "inter-procedural" analysis, meaning it can trace data flow across multiple functions and files to find bugs that are invisible to local checks. Using it is surprisingly simple—you don't need to change your code. Just prefix your usual build command with infer run -- (for example: infer run -- ./gradlew build or infer run -- make). Infer captures the compilation command, translates it, and outputs a list of potential issues directly in your console and in a generated infer-out/ directory. It is used at scale at Meta to analyze code changes every day. Learn more at https://fbinfer.com/ and read the engineering deep dive here: https://lnkd.in/epJ2kEkH

⚡️OSS Project Spotlight⚡️ IGListKit is a data-driven UICollectionView framework for building fast and flexible lists on iOS. It features a decoupled diffing algorithm that handles updates automatically, completely eliminating the need for reloadData or performBatchUpdates. Learn more: https://lnkd.in/gr5ruArj

📊 Curious how Python typing is shaping developer workflows this year? The 2025 Python Typing Survey uncovers why typing has become such a key part of Python development — and what parts of the ecosystem users want to see improved. Whether you’re a seasoned Pythonista or just exploring type hints, this is a must-read: https://lnkd.in/e4cnv_A5 Engineering at Meta

🔎 Check out our new video on Pyrefly’s Pydantic support. Pyrefly currently offers experimental support for Pydantic, bringing improved static type checking and IDE integration to your data models. Learn what's supported and how to start using Pyrefly with your Pydantic code today! 📹 Watch the video: https://lnkd.in/ggmW8NBV #Pyrefly #Pydantic #Python #TypeChecking #OpenSource

🚀 New from Meta AI Research: Souper-Model! By smartly averaging multiple model weights using a method called SoCE (Soup Of Category Experts), the team achieved strong performance across multilingual, tool-calling, and math tasks — without retraining a whole new model. 🔍 Key insight: benchmarks break down into categories that poorly correlate — so pick the “expert” model for each category and weight them non-uniformly for best gains. 📈 The result: meaningful boosts on major benchmarks. Read more here: https://lnkd.in/gtyErEhS #AI #LLM #ModelSoup #MetaAI #MachineLearning