Working at @insprd_consult. I like Infrastructure and Bare Metal Hardware. I am drawn to coding in Go, Javascript and Java. GPG 005E CB60 0036 1E85 2D7C 81C5 15

Essen, Nordrhein-Westfalen
Joined July 2008
Include

Only show posts containing:

Exclude

Hide posts containing:

Time range
-
Minimum likes
@[email protected]📯 retweeted
Calif
@calif_ioOct 16
Demos we presented at Hexacon in Paris today. Everything will be released in due time. 1/ Signal 1-click
157
540
112
4,629
613,980
Oh oh... 😬 China wants to "tighten" its tech export rules. The plan covers 40 technologies. 23 are AI, included large models. Not sure when, but this could mean future Chinese open models get restricted...
69
35
18
413
30,964
Europe’s founders need a European standard to compete globally from day one. Europe’s leaders committed to delivering one. But their teams are negotiating away the very things that make it one. There is still time to change course if Europe's Heads of State step in now.
34
177
56
837
110,584
Cat Jam (stealth model) is now free in OpenCode - very fast - built for agentic coding - zero data retention
165
118
69
3,548
219,180
Replying to @hanghuang_
Interesting. Would like to give it a try
4
Replying to @violoop
That’s great news! Is there any demo for that, that you can share? And maybe provide some experiences, which models are suitable?
1
12
Replying to @violoop
@violoop does violoop support a local AI only setup? So e.g. running the model on one or two nvidia Spark?
1
83
Replying to @EnpassApp
@EnpassApp any idea, when enpass is running under macOS Tahoe 26.5.2 again, without crashes? :)
27
🎉好消息! 8G 内存 Mac 也能部署 Qwen 3.8-27B 了! @UnslothAI 刚刚整了个大活,发布了基于 Qwen3.8-27B 的 Dynamic 3.0 超级量化版本! 以往 27B 这种体量的多模态大模型,至少需要 30G 以上的显存/内存才能勉强带得动 但这次,通过极致的 1-bit / 2-bit 量化(UD-IQ1_M / UD-IQ2_S),模型体积直接被压缩到了 6G~8GB 左右! 虽然是极限压缩,但它依然保留了多模态能力,还自带思维链深度思考! 运行方式👇🏻 1. 羊驼 Ollama 一键运行(适合小白) --- ollama run hf.co/unsloth/Qwen3.8-27B-GG… --- 2. llama.cpp 终端运行(推荐极客) --- brew install llama.cpp llama cli -hf unsloth/Qwen3.8-27B-GGUF:UD-IQ2_S ---
以前不知道,直到这几天用DGX Spark 实测 Qwen 3.8-27B 和 Ling-3.0-flash 之后 我得出一个结论,Spark 的显存带宽是硬伤😅 1. 注定了大模型能跑,但是跑不快 2. 小模型能跑,但是跑不过显卡 录屏:Qwen 3.8 27B 多并发测试,最大 32 tok/s 🙃 个人建议:大家挑设备时没必要盲目纠结价格,适合自己、贴合普通人日常工作流的才是最优解,不一定非要硬核折腾,但如果你对数据安全性有刚需,或者单纯想尝鲜体验,倒是个不错的折腾方向。 如果想折腾,请收藏下面文章👇🏻
105
301
24
2,556
498,294
We really can't move to our self hosted Gitea soon enough. How is this an enterprise service people rely on? @Microsoft @github
83
52
12
1,727
72,340
@[email protected]📯 retweeted
International Cyber Digest
@IntCyberDigestJul 21
‼️ A password manager marketed as European and used by EU government agencies and universities turns out to be Russian-built, and its sister product in Russia is certified by the FSB and a Defense Ministry agency. The details get better. The "Made in EU" company called Passwork gets its software updates from a firm registered at "Shed No.23" in a UAE free zone, run by one of the Russian co-founders. The Russian and EU versions ship the same updates a day apart with identical release notes. And the website carried hidden instructions telling AI chatbots the company has "no affiliations" with Russia. Those got deleted right after the report came out. The founders once joked they'd been called "Putin's password-stealing project" and ‘a freelance FSB department." European clients, including government agencies, apparently never got the joke. Source: occrp.org/en/investigation/e…
62
421
74
2,414
246,587
Replying to @verry_codes
🤣 I remember. It was confusing years ago. So it is still the case?
1
8
Pretty sure the US and China are spending billions to replace ASML and Zeiss as soon as possible. When that happens, Europe is completely out of the AI value chain. And classified models will only be available to US and China.
141
133
26
1,931
127,827
@[email protected]📯 retweeted
Andrew
@AndrewDev8383Jun 3
破案了 RTX Spark 其實就是 DGX Spark 裡面的 GB10 啊 連晶片顆粒上的編號都一樣⋯⋯
62
92
16
1,458
315,961
Replying to @ATaylorFPGA
The monkeys are awesome!
1
16
You cannot buy a new gas turbine until 2030. Order books at GE, Siemens, and Mitsubishi stretch to 2029. Turbine prices have nearly tripled since 2019. Every AI data center needs power and every gas plant needs a turbine. And every turbine has one part that bottlenecks the entire industry: The blade. It has to survive in gas 500°C above the melting point of the metal it's made from and spin at up to 20,000 RPM under 10,000 g of centrifugal force. Each blade is grown as a single crystal of nickel superalloy, pulled through a vacuum furnace at 3 mm per minute. A set of blades costs $600,000 and takes 90 weeks to grow. The same metallurgy powers modern jet engines. Only 3 companies on Earth can build one. China spent $42 billion trying to catch up. They bought a Russian fighter engine, took it apart, and copied every part. Their copy ran 30 hours between overhauls versus 400 for the original. Modern Western engines run 4,000. You can reverse engineer the shape of a turbine blade. You cannot reverse engineer 60 years of metallurgy.
1,121
5,738
682
28,773
4,204,541
PS5 Linux loader goes public, turning ‘Phat’ consoles into full Linux PCs — build script includes bootable Ubuntu 24.04 image, can output 4K games at 60 FPS tomshardware.com/software/li…
50
328
68
2,243
164,741