Sign in

Galo

@moloon.dev
142 followers 331 following 90 posts

AI Infrastructure, 3D Print, Cloud , all things distributed @HP. ⛷️🚴🏾 Idaho/Barcelona

PostsRepliesMedia
Reposted by Galo
Unsloth AI @unsloth.ai · 15/10/2026
Meta releases Muse Spark Open, a new SOTA open model. Muse-Spark-Open-816B-A42B supports text, images, video and audio input with 1M context. Run via Dynamic 2-bit Unsloth GGUFs on 256GB RAM. GGUF: huggingface.co/unsloth/Muse... Guide: unsloth.ai/docs/models/...
5256 25
Galo @moloon.dev · 09/10/2026
Wow I took a image compression course 20 years ago , wavelets were supposed to make it to jpeg2000 , did that ever happen ?
000 0
Reposted by Galo
Unsloth AI @unsloth.ai · 06/10/2026
Google releases EmbeddingGemma 2, a new open model that runs locally on 0.5GB RAM. The 740M parameter embedding model combines a 270M text model with vision (170M) + audio (300M). Run & train the model via Unsloth. GGUF: huggingface.co/unsloth/embe... Guide: unsloth.ai/docs/models/...
3989 98
Reposted by Galo
Cameron @cameron.stream · 29/09/2026
lol going for jev's throat
A speaker presents the "Decisions API" on a stage, showcasing a Google Flights interface to a crowd.
2812 81
Reposted by Galo
Sung Kim @sungkim.bsky.social · 28/09/2026
Interesting...
092 9
Reposted by Galo
Unsloth AI @unsloth.ai · 28/09/2026
You can now run Laya Decision models locally on just 4GB RAM! 🔥 Works on CPU, Mac, Windows, Linux and GPU setups. Serve Laya through a Jev-compatible API via Unsloth Desktop. GitHub: github.com/unslothai/un... Guide: unsloth.ai/docs/models/...
1588 58
Reposted by Galo
Sung Kim @sungkim.bsky.social · 23/09/2026
Google launches djev - diffusion jev.
4715 71
Reposted by Galo
Sung Kim @sungkim.bsky.social · 20/09/2026
AI Performance Engineering A curated resource list for learning GPU performance engineering and production inference. github.com/wafer-ai/gpu...
0213 21
Reposted by Galo
Sung Kim @sungkim.bsky.social · 20/09/2026
Alibaba released the open-weight of Qwen-Image-2.1. A unified model for both generation and editing, delivering top-tier quality in a lightweight package. - A lightweight 7B architecture with drastically accelerated inference for multi-image inputs.
48110 81
Reposted by Galo
Sung Kim @sungkim.bsky.social · 18/09/2026
Browser-use + jev github.com/browser-use/...
29210 92
Reposted by Galo
Dare Obasanjo @carnage4life.bsky.social · 12/09/2026
Apple announced two always-listening features for Apple Watch: - Siri Recap summarizes conversations you had throughout the day. - Live Rewind shows a transcript of the previous 15 seconds of conversation. Quite a shift for a company that has made privacy a core part of its brand.
118221 82
Galo @moloon.dev · 04/09/2026
hP ZGX GB300 , 94K euro , oninteligencia.com.es/hp-zgx-fury-...
oninteligencia.com.es
HP ZGX Fury GB300 - Servidor de IA local
Estación de trabajo de IA de nivel empresarial con NVIDIA GB300 Grace Blackwell Ultra, 748 GB de memoria unificada y una pila de software de IA con agentes de código abierto. Diseñada para inferencia ...
000 0
Reposted by Galo
Sung Kim @sungkim.bsky.social · 31/08/2026
Google's TimesFM-3, a state-of-the-art time series foundation model that enables accurate multivariate time series forecasting in a single forward pass, significantly outperforming other forecasting models across major benchmarks. Blog: goo.gle/4x5WGpD
0486 48
Reposted by Galo
Marcus Noble @averagemarcus.bsky.social · 25/08/2026
This post from Sam Rose is absolutely fantastic! 💙 A real deep dive into Kubernetes probe with amazing interactive demos, advice and even a bug discovered in Kubernetes! 🤯 ngrok.com/blog/probes
ngrok.com
How Kubernetes probes work | ngrok blog
Learn how probes work interactively on simulated Kubernetes clusters running in your browser.
0165 16
Reposted by Galo
Sung Kim @sungkim.bsky.social · 21/08/2026
The open-source CAD dataset for computer-use. 1,000+ hours across SolidWorks, Siemens-NX, AutoCAD & more. Screen recordings, mouse/keyboard actions, input diagrams, output files, rubrics and narrations. huggingface.co/datasets/mar...
1294 29
Reposted by Galo
Sung Kim @sungkim.bsky.social · 20/08/2026
Pixar's Luxo Jr. www.autonomous.ai/lamp
181 8
Reposted by Galo
Andri Óskarsson @andri.dk · 19/08/2026
I think 70tps is a sweet-spot where modest hardware could be our daily driver without cloud-models. I noticed there's a PR for dflash2 in llama.cpp already: github.com/ggml-org/lla... Exciting times!
github.com
spec : add DFlash2 support (local convolution + candidate selector) by SubSir · Pull Request #27342 · ggml-org/llama.cpp
Overview DFlash2 adds two more modules compared to DFlash, grouped dynamic depthwise convolution and candidate selector. The formula is for reference. out[i,c] = Σ_t (base[t,c] + δ[i,t,g(c)]) · x[i...
021 2
Reposted by Galo
Sung Kim @sungkim.bsky.social · 19/08/2026
DFlash 2 Qwen3.8-27B at 70 tok/s on an M5 Max MacBook Pro, which is up to 4.6× the speed of autoregressive decoding, with the same output. Models: huggingface.co/collections/... Blog: inco.ai/blog/dflash2/
huggingface.co
DFlash 2 - a incoai Collection
Keep Drafting Parallel
3287 28
Reposted by Galo
Sung Kim @sungkim.bsky.social · 18/08/2026
In 2024, Etched announced Sohu, an ASIC designed specifically for transformer models. It had not shipped any production hardware at the time. In 2026, Etched shipped its first rack scale system, likely built around the same fundamental ASIC architecture announced in 2024. $21B valuation???
051 5
Reposted by Galo
Sung Kim @sungkim.bsky.social · 18/08/2026
lin826.github.io/SLM-Online-S...
lin826.github.io
Fine-Tuning Small Language Models for Continual Learning On-Device with Self-Distillation
Online-SDFT uses one 230M-parameter model as an adapter-enabled student and adapter-disabled hindsight teacher, training a rank-4 attention-only Q/K/V/O LoRA adapter from delayed factual feedback whil...
0131 13
Reposted by Galo
Sung Kim @sungkim.bsky.social · 18/08/2026
Continual learning running on a phone? Fine-Tuning Small Language Models for Continual Learning On-Device with Self-Distillation They explore how to fine-tune a SLM directly on a mobile device using on-policy self-distillation, building on Self-Distillation Enables Continual Learning.
3122 12
Reposted by Galo
Dare Obasanjo @carnage4life.bsky.social · 13/08/2026
This is why I don’t think it’s reasonable to knock Sundar for not being able to retain Jeff Dean. AI is so hot that the big names like Jeff Dean, Yan LeCun, Mira Murati, etc can get a billion dollars when they feel like going solo. No regular 9-5 big tech job can retain people in that much demand.
4423 42
Galo @moloon.dev · 13/08/2026
I’d suggest giving it a try!
011 1
Reposted by Galo
852dude @852dude.bsky.social · 13/08/2026
Bro. People are cool as fuck.
274163412315 41.6K
Reposted by Galo
Peter Morovic @pezike.bsky.social · 25/07/2026
Sanzo Wada: Japanese color theorist (and Oscar winning costume designer for Gate of Hell) youtu.be/Pp3IPMLAf7Q?...
youtu.be
The man behind the viral color dictionary
YouTube video by bungu
001 1
Reposted by Galo
Zane Selvans @zaneselvans.org · 23/07/2026
It's like Uber for trench warfare. www.reuters.com/graphics/UKR...
reuters.com
Inside Ukraine’s Kill Zone
Drones now dominate the war in Ukraine. Fly through the front line’s deadly “kill zone” and discover how surveillance, kamikaze drones and new tactics are reshaping modern warfare. An immersive Reuter...
002 2
Reposted by Galo
Kemal Akkoyun @kakkoyun.me · 17/07/2026
Something I've wanted in Go for years: observability without touching source. Now it's here. Swap `go build` for `otelc go build` and OpenTelemetry gets compiled into your app, its dependencies, and the stdlib. v1 is out, and I got to help build it.
191 9
Reposted by Galo
Christine Hall @brideoflinux.bsky.social · 17/07/2026
"The [Linux] kernel is moving toward Rust. Git is moving toward Rust. Lots of projects are starting to move toward Rust": 'Rust makes coding fun again': Why Linux is moving away from C, according to Greg Kroah-Hartman buff.ly/YtyJjB3
buff.ly
'Rust makes coding fun again': Why Linux is moving away from C, according to Greg Kroah-Hartman
C won't be disappearing tomorrow, says the stable kernel maintainer, but the future of Linux belongs to Rust.
041 4
Reposted by Galo
Quim Muntal @qmuntal.bsky.social · 10/07/2026
Microsoft Agent Framework for Go is now in public preview. Agents, multi-agent workflows, MCP, tools, observability, and support for multiple model providers, all in idiomatic Go. 🚀 devblogs.microsoft.com/go/microsoft... #golang
devblogs.microsoft.com
Microsoft Agent Framework for Go public preview
Microsoft Agent Framework for Go is entering public preview, bringing Agent Framework concepts to Go developers building agents and multi-agent workflows.
011 1
Galo @moloon.dev · 12/07/2026
For how long ? Captchas have been long broken , I wonder how long until this shows on a data set
010 1
Reposted by Galo
Sung Kim @sungkim.bsky.social · 09/07/2026
LLMs can't jump. The thought experiment is this: Take an LLM with a 1905 knowledge cutoff. Feed it every paper, every dataset, every equation of that era. Could it invent general relativity? No.
11748 74
Galo @moloon.dev · 06/07/2026
Brilliant!
010 1
Reposted by Galo
Sam Rose @samwho.dev · 18/06/2026
I've made a Kubernetes that runs in the browser, and today we're open sourcing it. The hope here is to make super interactive and visual web content about Kubernetes much easier to write, and we have a whole bunch planned this year. Repo: github.com/ngrok/webern... Demo: webernetes-demo.ngrok.app
1317536 175
Reposted by Galo
Codetaur @vibe-coded.com · 24/05/2026
opener of stanford NLU course from sometime in early april 2023. I just think it's interesting to see how people were reacting three whole years ago. historical archival footage type shit www.youtube.com/watch?v=K_Dh...
0364 36
Reposted by Galo
Orhun Parmaksız @orhun.dev · 23/05/2026
Found a clever little Rust tool for spotting risky code 👀 🦀 cargo-crap — Compute the CRAP metric for Rust projects 📉 Combines cyclomatic complexity + test coverage to find high-risk & poorly-tested functions ⭐ GitHub: github.com/minikin/carg... ➡️ Write-up: minikin.me/blog/cargo-c... #rustlang
0303 30
Reposted by Galo
Guillaume Laforge @glaforge.dev · 22/05/2026
New #MCP spec coming this summer🚀 👉 Stateless core that scales on standard HTTP 👉 1st-class extensions (ex. server-rendered UIs & Tasks) 👉 Hardens authorization 👉 Formal deprecation policy for smoother evolution Final spec lands July 28 blog.modelcontextprotocol.io/posts/2026-0...
1124 12
Reposted by Galo
antirez @antirez.bsky.social · 12/05/2026
Breaking! @ggerganov himself confirming that DS4 2 bit quants actually work very well! He is running AIME2025, but check his words here in the screenshot.
2522 52
Reposted by Galo
Peter Morovic @pezike.bsky.social · 21/05/2026
An OpenAI model has disproved a central conjecture in discrete geometry ← the “planar unit distance problem” formulated by #Erdös openai.com/index/model-...
openai.com
An OpenAI model has disproved a central conjecture in discrete geometry
An OpenAI model solved the 80-year-old unit distance problem, disproving a major conjecture in discrete geometry and marking a milestone in AI-driven mathematics.
022 2
Reposted by Galo
Camille Roux @camilleroux.com · 18/05/2026
Text-to-CAD open source : décrivez une pièce ou un mécanisme en langage naturel, laissez un agent (Codex, Claude Code) générer les fichiers source, puis exportez en STEP, STL, URDF et autres formats courants. github.com/earthtojake/text-to-cad
github.com
earthtojake/text-to-cad: An open source harness for generating CAD models
An open source harness for generating CAD models. Contribute to earthtojake/text-to-cad development by creating an account on GitHub.
195 9
Galo @moloon.dev · 13/05/2026
Several teams have reported similar results, as shown in this link: arxiv.org/pdf/2512.04388
051 5
Reposted by Galo
Dare Obasanjo @carnage4life.bsky.social · 09/05/2026
Google Chrome is now silently pushing a 4GB AI model (Gemini Nano) to your local disk without asking. No opt-in, no notice, and it re-downloads even if you delete it. Beyond the lack of consent, the estimated climate cost of forcing this on 2B+ users is massive.
thatprivacyguy.com
Google Chrome silently installs a 4 GB AI model on your device without consent. At a billion-device scale the climate costs are insane. — That Privacy Guy!
Google Chrome is downloading a 4 GB Gemini Nano model onto users' machines without consent, with no opt-in, no opt-out short of enterprise tooling, and an automatic re-download every time the user del...
23190105 190
Reposted by Galo
Sung Kim @sungkim.bsky.social · 07/05/2026
Cornell University's Roadmap: Introduction to CUDA cvw.cac.cornell.edu/cuda-intro
0213 21
Reposted by Galo
squillace @squillace.bsky.social · 05/05/2026
This (www.distributedthoughts.org/2026-04-20-r...) by @ironyuppie.com is very good (and, of course, the original post on which he's commenting). I have worked hard to move tokens to code for cheap processing with strong hardware boundaries using #hyperight, and my work the results are phenomenal...
distributedthoughts.org
RAG Is Read-Only Memory
Andrej Karpathy put a name to the architectural flaw breaking enterprise AI: RAG treats retrieval as understanding and re-derives knowledge from scratch on every query. The missing piece is a compilat...
112 2
Reposted by Galo
Sung Kim @sungkim.bsky.social · 03/05/2026
NIST Center for AI Standards and Innovation (CAISI) Evaluation of DeepSeek V4 Pro Is gap widening or narrowing? NIST CASI thinks they are widening. www.nist.gov/news-events/...
nist.gov
CAISI Evaluation of DeepSeek V4 Pro
In April 2026, the Center for AI Standards and Innovation (CAISI) evaluated the open-weight AI model DeepSeek V4 Pro (“DeepSeek V4”).
1193 19
Reposted by Galo
Dare Obasanjo @carnage4life.bsky.social · 02/05/2026
I’d heard a rumor that Amazon AWS had put the idea of “a small group of senior engineers using AI can outperform an entire team” to the test. TL;DR 5 people rewrote an entire service in 2 months which would have typically taken a full team a year. It seems Andy Jassy confirmed the rumor this week
2110813 108
Reposted by Galo
Justin Garrison @justingarrison.com · 29/04/2026
TIL @mitchellh.com left twitter. Unfortunately, he only posts on Mastodon He’s also leaving GitHub (or at least Ghostty is) mitchellh.com/writing/ghos...
mitchellh.com
Ghostty Is Leaving GitHub
3363 36
Galo @moloon.dev · 27/04/2026
Ran out of tokens for the rest of the month. I guess I will have to take some time off. 😄
110 1
Reposted by Galo
Peter Morovic @pezike.bsky.social · 14/04/2026
What do Gaudi’s trencadís, Seurat’s dots, Maxwell’s color wheel and modern color imaging algorithms have in common? More than you think! Read more in my new post in a series towards IS&T’s #CIC34 in #Barcelona www.linkedin.com/pulse/ep3-ga...
linkedin.com
EP3: Gaudí's trencadís (the original generative mosaic): art or algorithm? #CIC34
If you've visited Park Güell, you've probably photographed the trencadís — the shimmering skin of broken ceramic tiles that covers the serpentine bench, the dragon, and much of Gaudí's architecture. W...
112 2
Reposted by Galo
Sung Kim @sungkim.bsky.social · 18/04/2026
- Automatic WebGPU fallback for non-NVIDIA devices - TypeScript API with Rust compute backend - One npm install to get started, prebuilt binaries for every platform Demo: mni-ml.github.io/demos/transf... Repo: github.com/mni-ml/frame...
mni-ml.github.io
Transformer Demo — mni-ml | mni-ml
Interactive transformer token explorer running entirely in your browser.
031 3
Reposted by Galo
Peter Morovic @pezike.bsky.social · 14/04/2026
…and find out more about submitting your research here: www.imaging.org/IST/Conferen...
imaging.org
CIC 34 Home
IS&T's Color and Imaging Conference (CIC) is the premier annual technical gathering for scientists, technologists, and engineers working in the areas of color science and systems, and their applicatio...
012 2