Loading…
Topic
Open Source
568 posts about Open Source. Every summary links to the original.
- 568 posts
- 36 companies
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
Newest first
Posts
Imaginary on Fly, just One-click
Fly announces a one-click launcher for Imaginary, an open-source image service that is among Fly’s popular global image-service use cases. The launcher is intended to let users create a running, production-ready version without infrastructure by pasting button code, and Imaginary’s creators added it to the project’s GitHub README. The post says users can get started with operations including converting and resizing images faster, while scalable SVG buttons are designed to work on pages. It also presents one-click launchers as a possible funding channel through revenue sharing and invites other open-source projects to help develop a process for launching applications on Fly.
How to generate text: using different decoding methods for language generation with Transformers
Auto-regressive language generation relies heavily on decoding methods to convert predicted next-token probability distributions into coherent text. Greedy search selects the single most probable token at each timestep, but it frequently leads to repetitive phrasing and misses high-probability sequences hidden behind low-probability intermediate words. Beam search alleviates this issue by maintaining multiple parallel hypotheses and can be paired with n-gram penalties to reduce immediate repetitions. Sampling approaches, including top-K and top-p sampling, offer alternative paths that can yield more fluent, varied open-ended generation. Using the Hugging Face transformers library, practitioners can configure these strategies via parameters on the model generate function.
How to train a new language model from scratch using Transformers and Tokenizers
Recent library enhancements simplify pretraining custom language models from raw text corpora. A small 84-million-parameter RoBERTa architecture comprising six layers, twelve attention heads, and a hidden dimension of 768 was pretrained on a 3 GB Esperanto text dataset. Training began by fitting a 52,000-vocabulary byte-level Byte-pair encoding tokenizer, which natively handled Esperanto diacritics and reduced average sequence lengths by approximately 30 percent compared to GPT-2 tokenization. The model was trained using masked language modeling, evaluated using fill-mask pipelines, and subsequently fine-tuned for part-of-speech tagging via standard token classification scripts. Final model artifacts and configurations were uploaded for community reuse via pre-trained model interfaces.
Supporting Open Source: Our Donations for 2018
Kickstarter describes its 2018 effort to support the open-source ecosystem through financial contributions aligned with its PBC charter and stated values. After surveying engineers for nominations, the company split its yearly donation budget among two open-source projects, Babel and Webpack, and four organizations supporting learning or coding skills: Advent of Code, Rails Girls Summer of Code, Code Nation, and Code Cooperative. Babel converts ECMAScript 2015+ into backwards-compatible JavaScript, while Webpack bundles JavaScript used at Kickstarter. The other recipients support collaborative problem-solving, participation by women and non-binary individuals, technology pathways for students in under-resourced high schools, and technology education for formerly incarcerated people. The post also notes an engineer’s weekly Code Nation mentoring and invites engineers to join the team and help choose future recipients.
Director of Engineering, Lawrence Mandel Talks Road to Leadership, Growth, and Finding Balance.
Lawrence Mandel, a Director of Production Engineering at Shopify, describes a career that moved from software development into technical leadership and the lessons behind that transition. He leads Developer Acceleration, a team that builds tools, conventions, and documentation so Shopify’s development organization can scale effectively. His path included web services, XML, open source, release engineering, project and release management, and self-taught Perl and CGI programming during a work term at IBM. Mandel says leadership required accepting that he could not know every technology or always be the most knowledgeable person in the room, while trusting others’ expertise. He also says prioritizing family and health before work improved his relationships and made his work more deliberate and efficient.
Handling Addresses from All Around the World
International address handling is difficult because countries and locales differ in field order, required fields, terminology, translations, and display conventions. Shopify models each address with fields such as names, address lines, zones, postal code, city, country code, and phone, then stores country-specific metadata in YAML-backed country data. A reusable format describes form and display order, allowing clients to reorder fields, prepopulate country and zone values, and add locale-specific characters such as Japan’s 〒 and 様. Label keys and translations handle terms such as province, prefecture, postcode, and pincode, while an API endpoint keeps applications synchronized and accepts locale information. The team also provides @shopify/address and higher-level components, while identifying caching as a remaining issue to reduce repeated network calls.
Fabric for Open Source Android
Open-source Android applications must remain buildable for contributors without exposing the credentials used by third-party services. Kickstarter’s Android application added Crashlytics through Fabric for crash reporting and real-time analytics, but integrating it required protecting the Fabric API key and secret from public source control. The post presents two normal SDK configuration paths—manifest metadata or fabric.properties—then uses a generated properties file, an example fallback, and Gradle logic to supply values from environment variables when available. Contributors without access to the secret repository receive an example apiKey value of 0, allowing the Fabric SDK to compile without the invalid null-key error, while the project’s real credentials remain outside the open-source codebase.
Heroku Proxying becomes Free Software
Heroku’s HTTP routing stack has three layers: state synchronization, request routing, and HTTP proxying. The platform team open-sourced the proxying layer as Vegur, an Erlang library that validates, normalizes, and forwards requests between clients and dynos. Unlike a conventional nginx reverse proxy, Vegur’s hosts and configuration are fully programmable, allowing state from hardcoded values, files, databases, or direct updates, while routing algorithms and debugging headers can also remain dynamic. Its delegated capabilities include domain lookup, backend selection, interface choice, resource checkout and checkin, customizable failure pages, per-application or per-backend controls for tunneling 100-continue requests, and upstream or downstream header insertion. Vegur had run in production on Heroku for over a year and powered the Heroku Private Spaces router.
Related topics
All topics →Weekly digest
The best engineering reads, once a week.