Today we’re introducing Gemma 4 12B — our latest open model that brings advanced agentic reasoning, vision and audio directly to your laptop.
It delivers performance nearing our larger Gemma models with a much smaller total memory footprint, while being small enough to run locally with just 16GB of VRAM. It’s open and accessible for everyone to use under a permissive Apache 2.0 license.
This is all made possible by our new, unified architecture that removes separate multimodal encoders. Here’s how we did it 🧵
@dangreenheck@juleshatton I mean … a thing for the Web that can’t be published? That license is weird … I understand the source code part but if my site wants to use this, can it use it? Do you provide the bundler too for public usage?
@dangreenheck@juleshatton that would happen as soon as somebody buys it and publish something on the Web though … it’s a weird position to state, it scares users too 🥲
Ported Google's Draco decoder to pure JavaScript.
https://t.co/jbXeB6AzsF
4.3× smaller than the WASM build, byte-for-byte identical output, often faster once you factor in load, init and parse.
@liamsmith86@ThePrimeagen Spark OS for DGS Spark is Ubuntu Desktop on ARM and it has pretty much anything you like, including Steam.
The only "blocker" for that platform is that it doesn't work best on Wayland, so it's a bit behind, but for AI work that doesn't matter at all.
x64 emulation also works ok
Defending electron is the most confidently wrong opinion in tech. Certainly wasnt on my bingo card.
Electron apps are bloated, slow, and take insane amounts of RAM to run.
The web framing is indicative of bad quality.
When I work with customers, I tell them “you can either have this correctly done, or you can have this quickly done”.
“Quick now, deal with the problems later” is usually the response, but the issues never get resolved and pile up.
Thats also why ai slop is slop, the quality of code models trained on was low, and they took the average of that.
What do you expect the quality to be, then?
@kenwheeler they probably have some HW with any NPU whatever in there, they had ti sell it as AI driven even if from "turn-on" to "change the volume of this BS" takes 30 seconds due badly configured bootstrap priorities 🤷
new in DS4 chat:
* bootstrap with your own system prompt
* highlight out of the box for code
* auto scrolling when images or other transformations happen as answer, right after the answer ends (load event for images)
https://t.co/TtGN4UvBfo
hey @antirez maybe you like this too, it's about our latest interaction around CORS and DS4 and I'm exploring the possibility to make it use the agent too, but that's for the next time 👋
@AdamRackis@valley_road4x In Italy they give limoncello (or other kind of liqueur, home made in some place) to you for free but who am I to change your idea about how to tip anyone if you’re that happy 😁
❗️ Over 30 official Red Hat npm packages were compromised. How they got in:
- A Red Hat employee's GitHub account was compromised.
- Attackers pushed "orphan commits" (detached from branch history) straight in, bypassing code review with no pull request.
- Payload "Miasma" (Mini Shai-Hulud variant) steals GitHub/cloud/Vault/SSH/npm secrets. Rotate everything since June 1.
- The commits added a workflow (ci.yaml) + script (_index.js) that abused npm trusted publishing, requesting a real OIDC token to publish backdoored versions.
@hthetiot Companies prefer Apache 2 over MIT (for more enforcement around the copyright inclusion, I suppose) … but licensing is hard, although companies rarely consider more restrictive licenses when considering adoption (specially corporates … they want to patent over OSS software 🤷)
that's not flat https://t.co/Q68t5CAMOn ... objects in arrays won't result flat.
for real flat results there's flatted-view and not much else around https://t.co/XVtzut4DWB
@satanacchio I want that to work for JS files too because that’s a TC39 proposal but no … my main bummer around that, I’d go Perry at that point (but it’s not there yet as replacement)
Hermes Agent comes with a truly absurd number of skills pre-enabled. Over 100 of them. This is roughly half.
I get what they're going for - they want an agent that comes "ready out of the box".
I just don't get why every user has to have a polymarket skill, 3 baoyu art skills (? never heard of this), a headless Pokemon skill, and Minecraft modpack server skills, all available the first time they run it.
I guess Hermes Agent just isn't for me.
@alexlomanto@jnardiello@antirez I feel you (I’m from Ancona though) but the fact Italians need to *fight* to improve Italy is, I believe, the whole elephant in the room or the entire reason most people migrate elsewhere, as sad as it is as reality