Remarkable progress of the Hugging Face science team in 2025: . Open-R1, smolagents, SmolVLM2, Ultra-Scale Playbook, OlympicCoder, Open Computer Agent, Reachy Mini, SmolVLA, LeRobot Hackathon and many more. A summary of the projects we released so far this year🧶

oh，smolvlm2 is good multimodal small model

I spent the whole week working on Basketball AI. - player detection and tracking with SAM2.- jersey number detection using RF-DETR.- number OCR with SmolVLM2.- team clustering with timm, UMAP, and KMeans. sports repo:

This robot works in real time locally running multiple AI models on apple MLX. Whisper for listening, smolvlm2 for image description (you can use Gemma for this but it adds so much latency), Gemma 3 4b for logic and autonomous action! Let me know if you want to check out the repo

we just dropped SmolVLM2: world's smollest video models in 256M, 500M and 2.2B ⏯️🤗. we also release the following 🔥.> an iPhone app (runs on 500M model in MLX).> integration with VLC for segmentation of descriptions (2.2B).> a highlights extractor (2.2B)

SmolVLM2

You can now train models with @huggingface's SmolVLM2 and SmolVLM 256M architectures on Roboflow 📈. Learn how to train a SmolVLM2 model for document processing in our latest guide 👇. #visionai #smolvlm #huggingface #computervision.

jersey numbers detection and OCR. - detect jersey numbers in real-time with RF-DETR .- crop number regions .- OCR the numbers with SmolVLM2

icymi I shipped a tutorial on fine-tuning vision language models on videos ⏯️. learn how to fine-tune SmolVLM2 on Video Feedback dataset 📖

Miquel and team COOKED: SmolVLM2 - Apache 2.0 licensed VideoLMs (ranging from 2.2B to 256M) - can even run on a FREE colab🔥. Beats models 5x it's size - runs on an iPhone!. Trained with memory efficiency as a focus you can pass extremely long videos with little VRAM required🤗

🚨🚨🚨SmolVLM2 is here - and it's a tiny titan! .This nano-sized model crushes image and video perception👁️🧠, all while being small enough to run on your iPhone, bringing cutting-edge multimodal AI to every device📲. No more cloud dependence! Your data is yours! #MobileAI

💥 HuggingSnap in the Apple Store! The easiest way to run SmolVLM2 500M in your iPhone 🤩

Exciting breakthroughs that dropped recently:. • Xbox revealed Muse, designed for gameplay ideation. • MatAnyone: Instant green screen for any video. • SmolVLM2: Tiny but powerful video understanding. • MagicArticulate: Auto-rigs 3D models for animation. • Animate

HuggingSnap is available on the App Store! 🥳. Turn your iPhone into a visual assistant that works entirely offline, ask it anything about what your camera is seeing, be amazed 👀. Free, private, open source - get inspired to build your own apps!. Based on SmolVLM2 and MLX.

The MLX community is growing rapidly!. I tried running a video/image processing app with SmolVLM2, but I need to dive deeper into optimizing it for lower-end devices. It keeps crashing on my iPhone 11—probably due to memory issues. #buildinpublic

One of our coolest projects just got cooler - check out VLC + SmolVLM2 in action!! 🔥 Catch the @videolan team showing it live at MasterDevFrance this Wednesday (Mar 12)! 🎬

SmolVLM2: Hugging Face's tiny titan of video AI! 🎥🤖 With models as small as 256M parameters, it deciphers videos locally on phones & laptops—no cloud needed. Privacy meets power, unlocking a new era of on-device video magic. 📱✨

Someone should create an app that would charge users automatically if they miss their workout. Make it offline and secure by using on-device visual LLM like SmolVLM2.

📱HuggingSnap is an open-source virtual assistant for iOS, powered by SmolVLM2, a compact, open multimodal model. It lets users learn from the world in real time using the 📸 camera. 🔒📴 Runs fully on-device, keeping your interactions private, even offline! (check the image)

Great job @rudrankriyam integrating SmolVLM2, this is fantastic!