Topic
Everything filed under Open Source, newest first.
RSS · JSON · All topics
There is no posttraining data commons
David Hall of Open Athena argues that LLMs got lucky with a pretraining commons humanity built for other reasons (Wikipedia, open source, Common Crawl), but posttraining consumes tasks and environments rather than documents, and no such commons exists; one has to be built deliberately.
10 min · 2,227 words
Marelleo, a story box I built for my kids
Arnaud Aubry describes building Marelleo, an open-source ESP32 story box that plays an audio story when a child places a figurine on top, as a DIY alternative to closed, content-subscription players like Toniebox, Lunii and Yoto, learning electronics and 3D printing from scratch along the way.
22 min · 5,163 words
Introducing Timestral: Mistral's time-series foundation models
Mistral releases Timestral, its first family of time-series foundation models in four sizes from 5M to 209M parameters. The largest ranks first on GIFT-Eval, TIME and fev-bench, delivers probabilistic multivariate forecasts with past and known-future covariates, runs efficiently on CPU, and ships under a modified MIT licence.
6 min · 1,385 words
Day 0 support for Muse Spark Open on vLLM
The vLLM team explains its day-0 support for Meta's Muse Spark Open, an 816B-total, 42B-active MoE model with 1M context and multimodal inputs: usage, model architecture details, DSpark speculative decoding, NVIDIA, AMD and initial TPU serving, and reinforcement learning support in vLLM.
7 min · 1,561 words
Welcome Muse Spark Open: open-weight frontier multimodal mixture-of-experts model from Meta
Hugging Face's guide to Meta's Muse Spark Open, an 816B-parameter (42B active) open-weight multimodal mixture-of-experts model with a 1M-token context: architecture, perception and audio encoders, and how to run it with Transformers, llama.cpp, SGLang, vLLM and MLX, deploy it, fine-tune it with TRL, plus demos.
19 min · 4,307 words
Ryan Dahl announced last week that the Deno team is joining Cloudflare. The Deno runtime winds down over the next year, JSR keeps running on Cloudflare's infrastructure, & the team's new job, per Cloudflare's post, is merging Deno's celld into workerd. The announcement came two years to the day after Deno 2 shipped. Bun joined Anthropic ten months ago. The tempting conclusion is that the alternative-runtime experiment is over & Node has won by default. In reality, the ecosystem has gotten the benefits Deno was pushing, just not from Deno itself.
6 min · 1,305 words
Comparing Chromium Development at Google and Igalia
Sharon Yang compares building Chromium inside Google versus at Igalia: process differences, review culture, release cadence, and what changes when the browser is your product versus one of many engines.
13 min · 2,902 words
The Rust team released 1.99.1, a point release fixing regressions from 1.99.0. The post lists the notable fixes and points users at rustup update as the upgrade path.
1 min · 321 words
Forgejo announces version 17.0 of the community-driven Git forge, covering notable changes, upgrade notes and release artifacts for self-hosted teams that want an open alternative to proprietary code hosts.
8 min · 1,863 words
Introducing Muse Spark Open, The Most Capable US Open-Weight Model to Date
Meta Superintelligence Labs releases the weights of Muse Spark Open, a natively multimodal 816B-parameter (42B active) model with a 1M-token context window built for coding and agentic work. It runs on two Apple M5 Ultra Mac Studios via MLX, is offered through hosted inference partners and major model hubs, ships with Helion GPU kernels for SGLang, and is released under the Muse Model License after safety evaluation.
3 min · 644 words
End-to-end encrypted relays for OCapN
Jessica Tallon of the Spritely Institute describes end-to-end encrypted relays for OCapN: how capability-secure networked objects can communicate through untrusted relays without exposing plaintext, and what that unlocks for peer connectivity.
12 min · 2,721 words
Virtual Oxide Labs: Emulating a Full Oxide Rack
Oxide's Product Assurance team introduces Voxel, a tool that emulates a full Oxide rack, including its real control plane, switches, service processors and customer-like networking topologies, so the team can test current software without scarce hardware; it covers image creation, launch, SP emulation and the MPL 2.0 release.
18 min · 4,170 words
Introducing the Accessible Components Library
TetraLogical's Craig Abbott introduces its open-sourced Accessible Components Library 1.0.0: eight components (accordion, breadcrumbs, dialog, disclosure, pagination, tabs, text input, textarea) with HTML, CSS and JavaScript, plus documentation of accessibility requirements, test scenarios, assistive technology compatibility and working automated tests, and a discussion of what automated tests can and cannot check.
8 min · 1,795 words
Felix Palmer announces A5 v1.0, the pentagonal Discrete Global Grid System that encodes any point, line or polygon on Earth as 64-bit cell IDs. The release freezes a permanent cell ID definition and adds compacted coverings with set operations, spatial hierarchy functions, equal-area cells and faster indexing across languages.
6 min · 1,313 words
On the 30th anniversary of KDE's original announcement, long-time contributor Cornelius Schumacher reflects on joining the project in 1999, the values of digital freedom it embodies, how its community keeps renewing itself, and the goals it set for enterprise, documentation and styling.
2 min · 462 words
Counting Stars: Open Source Hardware at Chip Companies
Zero ASIC founder Andreas Olofsson tallies GitHub stars on public chip design repositories by company to argue that the semiconductor industry contributes almost nothing to open source hardware. Ten-person Zero ASIC ranks first with about 3,900 stars, ahead of SiFive, NVIDIA, AMD and Intel, and the post explains the counting method and notes the work of startups and projects like OpenROAD.
6 min · 1,310 words
Making all divisions slower to fix a bug in 0.000000000000000000542% of cases
The Futhark compiler blog explains how signed integer division by -1 of the most negative two's-complement integer is undefined behaviour in the C code Futhark generates, how the compiler now adds a branch to every signed division to handle that single case correctly, what that costs on contrived and real benchmarks, and why unsigned division avoids the overhead.
8 min · 1,822 words
A new model for open source: closed to pull requests, open to remix
Geoffrey Huntley revisits The Cathedral and the Bazaar and argues that in the AI era open source needs a new model: keep code open but close it to pull requests, and instead publish guidance so people remix and patch software for their own needs, as he now plans to do with his own GitHub projects.
3 min · 680 words
Anthropic’s Scion is heavily inspired by Mononoke
A former Mononoke engineer compares Anthropic's description of 'derived data' in its new Scion version control system with how Meta's open-source Mononoke source control server used the same idea years earlier, and argues that Anthropic should have credited Mononoke as prior art.
4 min · 855 words
How the PHP Security Team Handles Your Reports
Alexandre Daubois of the PHP security team explains how security reports for php-src are triaged today, who does the work, the tools the team uses, what falls inside and outside the security policy, and how to send a report, including AI-assisted ones, that the team can act on.
5 min · 1,136 words