@xhinker

on Personal AI | open source contributor | startup founder member | ex-Microsoft blog: https://xitter.cf/t.co/QjnjnNVoCD

 WA, US
xhinker.medium.com/
Joined August 2009
  • Tweets3.1K
  • Following290
  • Followers314
490Media
Today, book "Using Stable Diffusion with Python" is published. Hope it is useful for SD developers. #stablediffusion #diffusion #ai amazon.com/Using-Stable-Diff…
7
1
10
2,894
This RTX 3090 unlock and boost up to 30% for free is real github.com/Zagitta/gpu-unloc… My unlock result:xhinker.medium.com/speed-up-…

GitHub - Zagitta/gpu-unlock: Unlock FP/BF16 instruction issue rate on Nvidia 3000's cards to their...

Unlock FP/BF16 instruction issue rate on Nvidia 3000's cards to their professional equivalents - Zagitta/gpu-unlock

github.com
2
140
xhinker.medium.com/strata-gu…

Strata-Guard: I Solved The Control Token Trap Nobody Else Does

Your model stops mid-sentence and the API says everything is fine. Sometimes it does the opposite: the thinking leaks into the answer. Both…

xhinker.medium.com
87
Back to the Qwen3.8-27B - 18 models tested I ran 18 different quantisations and inferences combinations on my Studio M5 Ultra, you should find this interesting... (more in the thread).
16
8
2
111
146,411
Human are flying with AI, everything is evolving so fast
22
v18 (2026-10-20) — single-window activation. Committing to a window now brings only that window to the front. Previously all your Chrome windows popped up above the app you were on. It now raises just that window github.com/xhinker/az-opt-ta…

GitHub - xhinker/az-opt-tab

Contribute to xhinker/az-opt-tab development by creating an account on GitHub.

github.com
32
Andrew Zhu retweeted
Ornith
@ornith_Oct 20
Aloha!🌺 Meet Ornith-1.5-122B: No Tradeoff Between Strong and Fast Stronger. Faster. Leaner. Ornith-1.5-122B delivers frontier-level capabilities with exceptional inference efficiency. With superior instruction following, stronger task decomposition, and deeper understanding of real-world workflows, it achieves state-of-the-art performance among open-source models of comparable size, while running on a single NVIDIA DGX card!! Open weights under the MIT license. Coding ✅SWE-bench Verified: 0.845 ✅TB2.1: 0.813 ✅CyberGym: 0.800 ✅NL2Repo: 0.512 Agentic ✅HLE: 0.446 ✅BrowseComp: 0.785 ✅MCP-Atlas: 0.746 ✅Toolathlon: 0.708 🤗Hugging Face: huggingface.co/ornith-ai/Orn… We can’t wait to see what you build with Ornith-1.5-122B.
222
523
459
2,776
1,493,014
Andrew Zhu retweeted
Niko Veit
@coldnikoOct 18
Last tests and builds are in progress. A small preview of what's coming in the new version. The release is scheduled for this late evening (European time). I'm open to giving answers to any of your questions below. And sorry for being late!
108
63
37
697
76,548
Those model release seems... mer..., Qwen4 are the real, and only real ones
1
2
253
One RTX 3090 + azllm:github.com/xhinker/azllm give you 262144 context and smart beast can build anything for you

GitHub - xhinker/azllm

Contribute to xhinker/azllm development by creating an account on GitHub.

github.com
1
32
oh, 100-250 tok/s stable
16
I Stopped Buying Software. My AI Agent Builds It for Me Now. xhinker.medium.com/i-stopped…

What Software Engineers Do When AI Build Softwares

I wanted one feature. The app gave me a thousand I’d never use. So I asked my AI to build just the one — and it made me rethink what…

xhinker.medium.com
1
54
Windows 11 style Option + Tab Window navigation for Mac: github.com/xhinker/az-opt-ta…
32
crazy good
AI Search
@aisearchioOct 15
Drama 3 by Fish Audio is now the top open TTS model. Allows meta tags within your transcript to control emotion, direction, SFX. Pretty good at voice cloning too. huggingface.co/fishaudio/dra…
22
Andrew Zhu retweeted
Niko Veit
@coldnikoOct 13
You can now run Qwen3.8-Flash-Next on any Mac Mini, Mac Studio or MacBook with Apple Silicon (Even without enough RAM). This is our base performance. From here, it will only get better.
28
31
23
279
87,745
Andrew Zhu retweeted
Beamsters
@Beamsters1Oct 13
Next Sushi 🍣 1.2.3 will upgrade Qwen Flash Next to have 20%+ less error with current GDN 16 bits, and with a new 32 bit default, quality will enhance even further. 32bit's tax is ~3% tokens/s, but quality is more important. Avoid that with a flag but still get 20% improvement.
12
2
59
2,347
TO save your life time, avoid Windows 11 as you can
1
42
wow
MATH FOURNITY
@math_fournityOct 8
🚨 卧槽无情！AI时代，行者无疆！ 黑苹果系统爱好者的福音，Nvidia显卡驱动来了！ github.com/nullmoth/nvidia-m…

GitHub - nullmoth/nvidia-macos-driver: Metal driver for NVIDIA GeForce cards on macOS Sequoia

Metal driver for NVIDIA GeForce cards on macOS Sequoia - nullmoth/nvidia-macos-driver

github.com
33
Great Great
Niko Veit
@coldnikoOct 8
Small update is out with lot’s of improvements and small fixes 👀 github.com/Niko1221/Strata/r…
30
crazy, love you
Niko Veit
@coldnikoOct 6
Replying to @LotusDecoder
Strata will support Qwen4 fully and there is still LOTS more to do. I can bring same performance as you now have on 5090 or 5070 with just 6GB VRAM. I’m only constrained with time and compute but will get better
38
Want repost it 100x times, try it out if you have a mac
Beamsters
@Beamsters1Oct 6
Sushi 🍣 v1.2 adds support for 0GB prefix-cache RAM to purely use ssd tier. This enables 48GB-64GB systems a lot more context for the same model at the cost of less than 1s waiting time. MLX-Serve's next release got this covered too.
26
Load more