Hugging Face

New Tutorial in the MCP Course! This tutorial walks you through building a tiny agents application using AMD hardware. 🔗 https://lnkd.in/efv9bk8H - You can use Neural Processing Unit (NPU) and integrated GPU (iGPU) from AMD - Sets up local file MCP for dealing with sensitive files It's sick to see local assistants make it on to even more hardware.

BOOOM! Both VLMs and LLMs now have a baked-in http server w/ OpenAI spec compatible API in transformers Launch it with `transformers serve` and connect your favorite apps. Here I'm running @OpenWebUI with local transformers. LLM, VLM, tool calling is in, STT & TTS coming soon!

I just tested Kimi-K2 with a benchmark I use for new models, and the results are 🔥🔥🔥🔥 The task: zero-shot binary classification of jailbreak prompts. > Best score across the board (97% accuracy vs 93% of Deepseek R1) > The second-fastest inference speed, thanks to Groq via Hugging Face, is 3x faster than R1. And the best thing? You can run similar experiments with no code and no install, thanks to Inference Providers and Hugging Face AISheets: https://lnkd.in/dVu2f2SB

Users of `torch.compile`. Some small performance tips: 1. Default to `fullgraph=True` to catch graph breaks as early as possible. 2. Check for recompilation triggers. Put your code under `torch._dynamo.config.patch(error_on_recompile=True)` context. 3. Use regional compilation almost always to cut down cold-start timing significantly. Graph-breaks and frequent recompilations can easily come in the way of performance. Eliminate them as much as possible. In Diffusers, we have a dedicated test suite for checking these things. Reference: https://lnkd.in/gK3DqscU

Kimi K2's speed and accuracy are mind-blowing 🤯 To put this power in the hands of AI builders, I'm excited to announce we just made it the default model for Hugging Face Sheets, powered by Groq. Time for some vibe testing! https://lnkd.in/dVu2f2SB Note: The video shows real-time SVG code generation

Kimi-K2 is ready for inference! In these notebooks I walk you through use cases like function calling and structured outputs. 🔗 https://lnkd.in/eTisJHEq You can swap it into any OpenAI compatible application via Inference Providers and get to work with an open source model.

one million thanks!

𝐂𝐨𝐧𝐟𝐮𝐬𝐞𝐝 𝐛𝐲 𝐌𝐂𝐏? Hugging Face is betting 𝐛𝐢𝐠 on it and you should too. Here are some resources so you don't miss out on a key piece of the AI puzzle: 📖 𝐅𝐚𝐬𝐭-𝐓𝐫𝐚𝐜𝐤 𝐘𝐨𝐮𝐫 𝐌𝐂𝐏 𝐄𝐱𝐩𝐞𝐫𝐭𝐢𝐬𝐞: My colleague Ben Burtenshaw and Anthropic have crafted a course that takes you from zero to expert in just a few hours. 🔗 https://lnkd.in/esWdd9dY 🛠️ 𝐑𝐞𝐚𝐥-𝐖𝐨𝐫𝐥𝐝 𝐌𝐂𝐏 𝐀𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧𝐬: Ready to apply what you learn? My guide on using Gradio MCP servers from Cursor provides a complete, hands-on example to solidify your understanding. 🔗 https://lnkd.in/exk4T3dB 🔬 𝐔𝐧𝐝𝐞𝐫 𝐭𝐡𝐞 𝐇𝐨𝐨𝐝: For engineers and curious minds, Shaun Smith's detailed write-up on constructing Hugging Face MCP servers demystifies the technology and gives some tips on deploying production-grade MCP servers at scale. 🔗 https://lnkd.in/eesg4JGu

Kimi K2 is number one trending on Hugging Face, just a day after release, congrats!

🚀 Deploy your own full-stack Desktop Agent in seconds! I’m excited to introduce ScreenEnv, a breakthrough python library that’s redefining what’s possible in desktop automation and AI agents. 💥 The problem we solved: Building desktop agents used to be a nightmare - clunky VMs, brittle scripts, expensive cloud setups. Most developers avoided GUI automation because… it just wasn’t worth the pain. 🧠 What ScreenEnv changes: 🖥️ Full desktop control Agents can now see, click, type, launch apps, manage windows, handle files and even record full sessions. 🤖 AI-Native by design Built for modern infrastructure with Model Context Protocol (MCP) support or a simple Sandbox API for smooth integration. 🐳 Instant setup Forget complex configs, just run it in docker. Get a full desktop environment up in less than 10 seconds. 🔒 Sandboxed & Safe Runs in isolated containers. Perfect for testing, deploying, or benchmarking without touching your host. The real breakthrough? You can now build agents that interact with applications, generate content, manage files, browse the web, and automate multi-step workflows all in real desktop environments. Who should care: ✅ QA teams — automate GUI testing with precision ✅ Businesses — AI agents that handle repetitive desktop tasks ✅ Developers — true cross-platform automation with zero boilerplate ✅ Researchers — reproducible, controlled environments for agent evaluation And yes — it’s built to plug seamlessly into your existing stack. 🙏 Huge thanks to my teammate Aymeric Roucher for their ideas, collaboration, and incredible energy during this release. ScreenEnv wouldn’t be what it is without you! 👉 ScreenEnv is ready to explore: https://lnkd.in/gC_yiuMT 📚 Check out the full blog post for more details: https://lnkd.in/gCHcWFC2 💬 What’s your biggest challenge in desktop automation right now? I’d love to hear your thoughts. Let’s build agents that actually work, not just in theory, but in the real world.