@NainsiDwiv50980 @karpathy Cool, except 1 token per second is unusable…
English
Fabio Cermelli
381 posts
Fabio Cermelli
@fcdl94
CTO and Cofounder of @FocoosAi. PhD in Computer Vision and Continual Learning at @PoliTOnews. Past president of IEEE @HKNPoliTo Mu Nu Chapter.
Piedmont, Italy Katılım Temmuz 2011
397 Takip Edilen640 Takipçiler
🚨 Someone just did the “impossible”…
They ran a ~400B parameter AI model on a laptop.
No cloud
No data center
Just a 48GB MacBook 🤯
A dev fed Claude Code with:
• @karpathy autoresearch repo
• Apple’s LLM in a Flash paper
• Goal: run Qwen3.5 397B locally
And it actually worked.
→ ~1 token/sec
→ ~21GB RAM
→ Rest streamed from SSD
This isn’t a flex
This is a shift
We’re entering a world where:
Your laptop can run models that once needed entire server farms
It’s not about more compute anymore
It’s about smarter systems 🚀
Suryansh Tiwari@Suryanshti777
English
@SergioPaniego @UnslothAI Yes, let's move it where it belongs. Thanks anyway!
English
@fcdl94 @UnslothAI I couldn't reproduce the error running that example with the latest trl (+transformers)😅 If you still find it, you can raise an issue in the repo with versions
English
Qwen3.5 dense (smol 🤏) models just dropped
- natively multimodal
- 0.8B · 2B · 4B · 9B (+ base variants)
- 262K context extensible to 1M
- built-in thinking
fine-tune them with TRL out of the box → SFT, GRPO, DPO and more!
English
@SergioPaniego Anyway, we found it's working well using @UnslothAI 🤷♂️
English
@SergioPaniego Of course!
File ../modeling_qwen3_5.py:1551, in Qwen3_5Model.get_rope_index(...)
IndexError: The shape of the mask [517] at index 0 does not match the shape of the indexed tensor [447] at index 0
We're using a fresh environment with the latest version, code is exactly yours
English
@MatznerJon That is really funny! May I ask what's your setup? What models are you using behind?
English
This is either going to be the best or worst idea I've ever had.
Hooked up my OpenClaw to all of our internet connected cameras at the house.
Got this this (OUT OF NOWHERE) this morning.
English
@sarahookr @adaptionlabs I hope you're going small, local and task-specific. That's the future of AI!
English
Beginnings are very special. Today is an important day for @adaptionlabs.
Today a handful of one-size-fits-all-models are optimized for the average use case.
Averages erase the exceptional.
Everything intelligent adapts. So should AI.
English
@liuziwei7 The 0.4B footprint is exactly what's needed for edge deployment. How does latency scale on edge hardware? What kind of hardware did you test?
English
🚤Real-Time Streaming VLA for Dynamic Manipulation🚤
#DynamicVLA is a 0.4B vision-language-action model that manipulates *moving* objects in real-time, with continuous inference and latent-aware action streaming
- Project: infinitescript.com/project/dynami…
- Code: github.com/hzxie/DynamicV…
AK@_akhaliq
DynamicVLA A Vision-Language-Action Model for Dynamic Object Manipulation
English
Yes, but how much are people spending for this? We need those models and agents to run locally in our machines to really achieve a new level for AI
Andrej Karpathy@karpathy
What's currently going on at @moltbook is genuinely the most incredible sci-fi takeoff-adjacent thing I have seen recently. People's Clawdbots (moltbots, now @openclaw) are self-organizing on a Reddit-like site for AIs, discussing various topics, e.g. even how to speak privately.
English
@NTWANtv @leadlagreport I love this question. That’s the real one. How much is people spending on this crap?!
English
@leadlagreport I like how contrarian you are. Moltbook is pure AI froth nonsense. How much compute is actually wasted for nothing more than an AI larp?
English
I’m telling you all right now that what’s happening with Moltbook just started the AI bear market.
If you aren’t wildly disturbed by what’s happening, you aren’t paying attention.
Few understand this.
English
Computer Vision is moving past the "box" era. 📦💀
We put a VLM on a Jetson Orin @NVIDIAAI to give it an actual brain. 🧠⚡️
No training. No labels. Just a prompt.
The VLM doesn't just alert; it reasons.
The future of video is contextual, explainable, and running on the Edge
English
@elonmusk @farzyness Actually, quite the opposite.
I know I can do it and I know how to do it.
Just not with the techniques everyone is currently betting on.
My bet is (famously) on JEPA, world models, and planning.
At some point, you'll realize I'm right 😅
English
Why is Yann such a Negative Nancy all the time?
The Humanoid Hub@TheHumanoidHub
Yann LeCun says absolutely none of the humanoid companies have any idea how to make those robots smart enough to be useful.
English
@giffmana I’m more worried about papers without code than with AI-wrote citations (especially if they are citiations of related work section)
English
@fcdl94 Just think about what steps an author needs to take for having this in their paper.
With that in mind, I don't think I'll trust their baselines to be proper either.
English
The NYU and Genentech authors used Word Copilot Instant or what?!
I think it's fine to use AI to help you write, but you need to stand behind every single word in your paper. Just like before AI. And it seems there's a lot of people who don't. Just like before AI...
Alex Cui@alexcdot
Okay so, we just found that over 50 papers published at @Neurips 2025 have AI hallucinations I don't think people realize how bad the slop is right now It's not just that researchers from @GoogleDeepMind, @Meta, @MIT, @Cambridge_Uni are using AI - they allowed LLMs to generate hallucinations in their papers and didn't notice at all. It's insane that these made it through peer review👇
English
@itsPaulAi @Cryptosaurus__ I’m using Qwen too. Is there any comparison?
English
@Cryptosaurus__ Waiting for the Unsloth version. But I'm already running Qwen 30B A3B (same size) so I'm not worried.
English
This is so good 🔥
You can run this new model on a laptop which is:
- 100% open source
- Only 3B active parameters (!!)
- Way better than GPT-OSS
- Perfect for vibe coding (and more)
And already available for free on Hugging Face or via API.
Open source models keep winning!
Z.ai@Zai_org
Introducing GLM-4.7-Flash: Your local coding and agentic assistant. Setting a new standard for the 30B class, GLM-4.7-Flash balances high performance with efficiency, making it the perfect lightweight deployment option. Beyond coding, it is also recommended for creative writing, translation, long-context tasks, and roleplay. Weights: huggingface.co/zai-org/GLM-4.… API: docs.z.ai/guides/overvie… - GLM-4.7-Flash: Free (1 concurrency) - GLM-4.7-FlashX: High-Speed and Affordable
English
#CVPR
What is the best use of the review title?
I struggle to find something which doesn't sound stupid and generic like "nice paper but it has drawbacks"
English
VLAs are getting traction. 2026 will be hot for robotics!
ModelScope@ModelScope2022
🤖 Introducing InternVLA-A1 — now fully open-sourced! Many VLA models follow instructions well in static scenes… but struggle in dynamic environments (conveyor belts, rotating platforms, multi-robot setups). Why? They see the present—but can’t imagine the future. InternVLA-A1 solution: unify perception, imagination, and action in one model: ✅ Scene understanding: Image + text → task parsing ✅ Task imagination: Predict future frames → reason about dynamics ✅ Guided control: Execute actions steered by visual foresight Powered by InternData-A1 - Large-scale high-quality simulated dataset, InternVLA-A1 stays robust under complex backgrounds, lighting, and distractions. 🔥 See it in action: 1️⃣ High-speed conveyor: track, predict, and stably grasp or flip packages 2️⃣ Rotating platform: task-aware recognition & precise pick-up of diverse items 📊 Outperforms π0 and Gr00t N1.5 on general manipulation benchmarks! ✨ Model, data, and code are all open! Models: modelscope.cn/models/InternR… Datasets: modelscope.cn/datasets/Inter… GitHub: github.com/InternRobotics…
English
I love this. I can only see it as a robotic chicken and I guess it’s the first of ita kind 🐓
Science girl@sciencegirl
KOU-III from Shandong University in China is a flying and walking robot with drone rotors and bipedal legs.
English