Jordan

Ai Note taking app - I install and use AI Locally on your devices for offline use. Mindsort.app is the first of many of my AI initiatives!

@HowToAI_ Game development is only going to get so much more wild with the tech as time goes on. I'm excited to see the future of new indie games

Microsoft has released a 4B parameter model that turns any image into a 3D asset in 3 seconds. It uses a new geometry format called O-Voxel that converts to a textured mesh in under 100ms on CUDA. Outputs GLB files with full PBR textures, ready for Blender, Unity, and Unreal. 100% Open Source.

@EnricoMagni1 @chesnyfcb Not even in the same league

@chesnyfcb How do the local models perform compare to Claude though?

Un tipo pagaba $200 al mes por Claude Max. Se fundió su suscripción en 3 horas de trabajo. Compró un Mac Mini básico por $599. Le instaló 5 modelos locales. Un comando. Un flag. Sus vecinos de oficina pensaban que estaba minando criptomonedas. Simplemente le enseñó a la máquina a clasificar mensajes, comprimir el contexto y mantener el sistema vivo mientras él dormía. A las 4 a.m., Claude alcanzó su límite de peticiones. El modelo local tomó el relevo. Por la mañana, leyó los logs: todo funcionó. Ni siquiera tuvo que despertarse. Un equipo haciendo lo mismo significa 3 ingenieros y $15.000 al mes en costes de API. Él pagó $599 una sola vez. 35 mil millones de parámetros en 16 gigas de memoria. Todos decían que era imposible. Un flag en un comando les demostró a todos que estaban equivocados. Y de personas como él... solo hay un puñado hasta ahora.

Drinking hydrogenized water every morning. Anything to help clean up the oxidative stress on the brain.

@RealProductGirl Welcome back. Here's to a productive and successful week 💪

And we back, fam ✨ New week, fresh energy, clean slate. Whatever you’re building, show up for it today. Have a beautiful start to your week 🤍

@RealProductGirl Bummer! Plumbing issues suck! Hopefully it all gets resolved quickly!

@HaareBlond @stevibe @TeksEdge You do realize parallel batch processing splits the bandwidth right?

@JordanDevAi @stevibe @TeksEdge The pic is real, but it's 21 requests batched together in vLLM. A single stream of Gemma 4 31B Dense on an RTX 5090 is roughly 50–90 tok/s, not 500+. And a 3090 on dense 30B is ~35 tok/s, not 300. The claim confuses total batched throughput with per-user speed

Qwen3.6 27B landed yesterday, so I ran it on 4 setups side-by-side to see how they stack up: 🔴 RTX 4090 — 45.59 tok/s, TTFT 525ms 🟢 RTX 5090 — 51.83 tok/s, TTFT 752ms ⚫️ M2 Ultra — 22.30 tok/s, TTFT 216ms 🟣 DGX Spark — 11.08 tok/s, TTFT 319ms This is a standard test: no tuning, just the out-of-the-box experience. For the NVIDIA cards I used llama.cpp with Unsloth's UD-Q4_K_XL quant. For the M2 Ultra I used MLX with Unsloth's UD-MLX-4bit quant, since MLX is the native path on Apple Silicon. Please consider this as the baseline, you can definitely squeeze more out of every one of these with fine-tuned settings.

@liquidai @MercedesBenz Congrats - that's awesome!

We’re entering a multi-year partnership with @MercedesBenz to scale embedded, on-device intelligence for their third- and fourth-generation MBUX. Our goal: to make the driver/vehicle relationship even more natural and effortless. Read more about our partnership: liquid.ai/press/liquid-a…

@stevibe @TeksEdge I have no problem getting 500-700 tok/s on Dense 30B models on my 5090 and 300 tok/sec on my 3090. Its not so much the quant as it is compiling Llama.cpp for your specific hardware. This pic is Gemma4 running their dense 31b at 500+ tok/sec.

@TeksEdge Not sure, I'm going to follow some guides and verify them.

@googlegemma And by both, I mean dual instances of Gemma4

@googlegemma I've been running both 24/7 for the last 6 days:

What does it take to run 3, 5, or even 10 concurrent instances of Gemma 4 locally? We've open-sourced a demo letting you run multiple models side-by-side on your hardware. Gemma 4 26B A4B easily runs 10+ concurrent requests on a MacBook Pro M4 Max at 18 tokens/sec per request.

It's efficient It does everything I need to do And its freedom of choice No bloat

@RealProductGirl The puppers 🐾

Does anyone remember their last night before a move? Was sleep non-existent? I’m getting the feeling I won’t get any 🥺

@BuescherScott Yo. That's wild. I recently formed a Real estate tech company with a partner who owns a realtor company in South Florida. We're publicly launching soon. I'll follow you back and check out your project.

@JordanDevAi Hit me up I have 3 real estate app ideas. I am a homebuilder in FL. Finishing up Cornerstonepm.ai rn. I need collab!

I have about 3 million photos I need to run classification on. Can someone do the AWQ 4 bit quantization for me

@RealProductGirl @BuescherScott Uh oh! As in... you need to move out of there like right now?! 🫣

@BuescherScott @JordanDevAi Bring a fan! My lights just went out 👀

Moving sucks...1 out of 5 stars...would not recommend Hope everyone has an awesome start to their Monday!

@0thernes_ai @Kimi_Moonshot On pure CPU with 0 GPU inference?

@JordanDevAi @Kimi_Moonshot I didnt train one im running right now and get 250+ doing nothing .

Come introduce yourself to the team, we have your slippers ready. Reach out at: talent@moonshot.ai

@RealProductGirl Like when I saw your post at 3am eastern time and I responded the other day lol

If I build it, they will come. So I keep grinding for every builder who's up late shipping, debugging, and refusing to quit. Your work matters. Keep Building. 🔨 I have you.

@petergyang If going for system ram, get as much as you can afford then use MoE and offload layers to CPU/GPU. Then you can run 80B+ MoE models

What is the sweet spot in open source model size? Are 35B models enough for local agentic workflows? Trying to decide how much RAM I need in a new computer.

@DavidOndrej1 Last week lol

if you're still running Gemma 4, you're falling behind imagine being so far behind the cutting-edge you're literally running LAST WEEK's AI model bro pack it up. you've already missed the AI revolution.