Hugging Face

Fine-tuning 5B param video models should be possible with a SINGLE 24GB GPU 🍓 We're releasing CogVideoX-Factory, a repository containing memory-optimized scripts to fine-tune Cog family of video models for T2V and I2V 🧪 Kudos to the hard work of Aryan V S with help from Yuxuan Zhang, and `glide-the` for making this happen. Go: https://lnkd.in/gY6JX8xw

🔒Gradio 5 Just Got Even More Secure!🔒 🛡️A comprehensive Security overhaul with Trail of Bits! Following the launch of Gradio 5, we are very excited to share about one of the most significant enhancements in Gradio 5 -- Web Security. With Gradio becoming the go-to framework for building and sharing machine learning apps (over 6M monthly downloads & 470K apps on Hugging Face Spaces), we knew that security had to be a top priority. That's why we went ahead and partnered with the cybersecurity experts at Trail of Bits for an in-depth security audit. 🔍 𝐈𝐧 𝐭𝐡𝐞 𝐬𝐩𝐢𝐫𝐢𝐭 𝐨𝐟 𝐭𝐫𝐚𝐧𝐬𝐩𝐚𝐫𝐞𝐧𝐜𝐲 𝐚𝐧𝐝 𝐨𝐩𝐞𝐧-𝐬𝐨𝐮𝐫𝐜𝐞, 𝐰𝐞'𝐫𝐞 𝐦𝐚𝐤𝐢𝐧𝐠 𝐭𝐡𝐞 𝐟𝐮𝐥𝐥 𝐬𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐫𝐞𝐩𝐨𝐫𝐭 𝐩𝐮𝐛𝐥𝐢𝐜! 📄 Keep reading to access the report and to learn more 👇 Gradio has successfully addressed all the issues identified by the Trail of Bits team in our security audit. The key highlights are: 1️⃣ Local Apps Security: Addressed CORS misconfigurations to prevent unauthorized access and token theft when running Gradio apps locally. 2️⃣ Deployed Apps Protection: Fixed SSRF vulnerabilities to secure internal networks and prevent data leaks for apps deployed on servers. 3️⃣ Shared Links Safety: Secured communication between frp-client and frp-server to prevent interception and unauthorized access when sharing apps via built-in share links. 4️⃣ Supply Chain Defense: Hardened our CI pipeline against supply chain threats by pinning dependencies and enhancing workflow security. 👉 Read the Full Security Report here: https://lnkd.in/djFewzS5 Q: What Does Gradio's Security Audit mean for You? 🤔 A: Your Gradio apps are now safer by default, without requiring significant changes to your code! You can continue building and sharing amazing ML apps with greater peace of mind. 🚀 💪 Upgrade to Gradio 5 Today with Full Confidence : pip install --upgrade gradio Dive Deeper: Check out our latest blog post on Hugging Face for an in-depth look at our security journey and the steps we've taken to make Gradio more secure: https://lnkd.in/dYcxMumQ We'd Love to Hear From our Community ! Your feedback is invaluable to us. Together, let's build a safer and more robust ecosystem for machine learning applications! 🙌 Let us know your thoughts on the new security enhancements with Gradio 5, or any other features you'd like to see in Gradio. 💬

Gradio 5 comes with Server Side Rendering (SSR) that makes loading of apps extremely fast. Upgrade today to start building your AI apps with the new state of the art! Read more about the release in our official blog on Hugging Face: https://lnkd.in/grnvQuzF

Fine-tuning 5B param video models should be possible with a SINGLE 24GB GPU 🍓 We're releasing CogVideoX-Factory, a repository containing memory-optimized scripts to fine-tune Cog family of video models for T2V and I2V 🧪 Kudos to the hard work of Aryan V S with help from Yuxuan Zhang, and `glide-the` for making this happen. Go: https://lnkd.in/gY6JX8xw

Using Hugging Face models and datasets is powerful for machine learning, but scaling tasks like model inference on large datasets can be challenging. Dask handles out-of-core computing, breaking up datasets into manageable chunks so that even large-scale tasks can run smoothly. In this example, we processed the FineWeb dataset (~715 GB in memory) using the 🤗 HF FineWeb-Edu classifier. Locally, processing 100 rows with pandas took ~10 seconds, but scaling up to 211M rows was possible with Dask on multi-GPU clusters deployed with Coiled. Results: - Handled large-scale text classification, filtering, and saved results to Hugging Face storage - Optimized GPU utilization to efficiently use expensive hardware This example could be adapted for other workflows like: - Genomic data filtering - Large-scale content extraction - Multimodal AI (audio, image, text) Had a lot of fun learning about Hugging Face and putting together this example with Quentin Lhoest, James Bourbeau, and Daniel van Strien Blog post: https://lnkd.in/gcV348fA

Let's go! First real good open-source text-to-video model with MIT license! Pyramid Flow SD3 is a 2B Diffusion Transformer (DiT) that can generate 10-second videos at 768p with 24fps! 🤯 🎥✨ TL;DR; 🎬 Can Generate 10-second videos at 768p/24FPS 🍹 2B parameter single unified Diffusion Transformer (DiT) 🖼️ Supports both text-to-video AND image-to-video 🧠 Uses Flow Matching for efficient training 💻 Two model variants: 384p (5s) and 768p (10s) 📼 example videos on project page 🛠️ Simple two-step implementation process 📚 MIT License and available on Hugging Face ✅ Trained only on open-source datasets 🔜 Training code coming soon! Model: https://lnkd.in/ee_hvaZU Paper: https://lnkd.in/eNhaySaX Project page: https://lnkd.in/eUj_4Fqf

Excited to welcome the first stable release of Gradio 5 today👋 Gradio has >2M users/month, >470K AI apps on Hugging Face Spaces. We have become the default way to build, share, and use machine learning applications among the AI community. Our goal with Gradio 5 was to address the most common pain points that we’ve heard from Gradio developers about taking these apps to production: 1️⃣ "Gradio apps load too slowly" Gradio 5 ships with major performance improvements, including the ability to serve Gradio apps via server-side rendering (SSR) which loads Gradio apps almost instantaneously in the browser. No more loading spinner! 🏎️💨 2️⃣ "Uh, this Gradio app looks old-school" Many of the core Gradio components, including Buttons, Tabs, Sliders, as well as the high-level chatbot interface, have been refreshed with a modern design in Gradio 5. We’re also releasing a new set of built-in themes, to let you easily create fresh-looking Gradio apps👨🎨 3️⃣ "I can’t build realtime apps in Gradio" We have changed how streaming works internally: in some cases, we now use websockets and send data via base64 to minimize latency. We support WebRTC via custom components, and have also added significantly more documentation and example demos that are focused on common streaming use cases, such as webcam-based object detection, video streaming, real-time speech transcription and generation, and conversational chatbots 🎤 4️⃣ "Are Gradio apps secure?" Gradio 5 intended to be a production-ready web framework for all kinds of machine learning applications, so we've also made significant improvements around security (including getting a 3rd-party audit of Gradio) -- more about that in an upcoming post! 5️⃣ "LLMs don't know Gradio" → Gradio 5 ships with an experimental AI Playground where you can use AI to generate or modify Gradio apps and preview the app right in your browser immediately: https://lnkd.in/gAChjNq4 🤖 👉 Try Gradio 5 today, its as free and intuitive as its always been: `pip install --upgrade gradio` Read all about Gradio 5 in our Release Blog: https://lnkd.in/grnvQuzF 🤩 We'd love to hear what you think!

Don't use an LLM when you can use a much cheaper model. I keep hearing this, and I agree (who can disagree with that?) The problem is that no one tells you how to actually do it. Just picking a pre-trained model (e.g., BERT) and throwing it at your problem won't work! If you want a small model to perform well on your problem, you need to fine-tune it. And to fine-tune it, you need data. The good news is that you don't need a lot of data but instead high quality data for your specific problem. This workshop with David Berenstein will show you all the steps to make 'use a smaller model' a reality for your next project. All with free and open-source technologies! Don't miss it!

[Cmd]+[Shift]+[Enter] ➡️ HuggingChat on Mac Game Changer! Kudos Cyril Zakka, MD at Hugging Face for building my favorite Mac app! Download link in the comments.

One million data labelling UIs for one million open models