Hugging Face

ModernBERT, BERT revisited in the age of LLMs and Generative AI! LightOn and Answer.ai modernized BERT! Improved architecture with 8192 context length, flash attention, and trained on 2T tokens. ModernBERT outperforms version BERT and RoBERTa versions! 👀 TL;DR; 2️⃣ Comes in 2 sizes base (139M) and large (395M) 🚀 Better performance across all metrics than the original BERT 📏 8,192 token context length (16x longer than BERT) ⚡ Modern architecture with Flash Attention 2, RoPE embeddings, and alternating attention 📚 Trained on 2 trillion tokens, primarily English and Code 💨 2-4x faster than other models with mixed-length inputs 🔓 Released under Apache 2.0 🤗 Available on Hugging Face and Transformers (main) Models: https://lnkd.in/ethiJ2xh Blog: https://lnkd.in/ebiEzb4P Paper: https://lnkd.in/ezR8MUBF

The only fine-tuning guide you need for 2025 ‼️ Excited to share “How to fine-tune open LLMs in 2025 with Hugging Face” covering everything from Q-LoRA to Spectrum methods with focus on optimization, efficiency and distributed training. 👀 Fine-tuning still matters for specialized use cases despite better models - especially for consistency, domain expertise, controlling output style, or reducing costs. In this guide you will learn how: 🎯 Define good use cases for fine-tuning vs. prompting 🛠️ Set up your development environment using Hugging Face libraries 📚 Create and prepare datasets in conversation format ⚡ Use Q-LoRA for efficient 4-bit training or Spectrum method to selectively fine-tune important layers 💨 Speed up training with Flash Attention and Liger Kernels 💻 Scale across multiple GPUs with DeepSpeed and accelerate 📊 Test and evaluate models with evaluation harness 🔥 Run for production using TGI/vLLM Blog: https://lnkd.in/egXtAwz5

NOW you can add AI to your Slack, Discord in just few steps with Gradio!🤩 🔥Create Slack apps, Discord bots, or Intercom-style website widgets in ANY modality (Text, image, Video, Audio, Omni etc)! Keep reading to learn how ⬇️ Guide: 🚀 Creating a Slack Bot from a Gradio App 🚀 Read here: https://lnkd.in/g2_Bydrj 🤎Do you love building stuff with Gradio? Support us on GitHub: Gradio.dev

Introducing FineWeb-C 🌐🎓, a community-built dataset for improving language models in ALL languages. Inspired by FineWeb-Edu the community is labelling the educational quality of texts for many languages. 318 annotators, 32K+ annotations, 12 languages - and growing!🌍

Fine-tune ColPali for your multimodal RAG use case 🔥 ColPali just landed to Hugging Face transformers and I have built a simple fine-tuning tutorial with QLoRA 🤗 You can fine-tune the model with 32 GB VRAM with batch size of 4 (which can run on Colab A100) Link in comments 💬

TIL: You can now load any Hugging Face dataset into PostgreSQL with just 1 line of SQL 🤯 All thanks to the pgai PostgreSQL extension. Shoutout to Matvey Arye from the Timescale AI engineering team for implementing this. Learn more about using PostgreSQL with HuggingFace datasets in the HuggingFace docs: https://lnkd.in/eS4hqSDq #postgresql #huggingface #opensource

🎢 Push to Hub: Export your dataset to the Hugging Face Hub directly from the Argilla UI. We’re super excited to announce that we've closed the loop: now you can load a dataset from the Hub, open it on Argilla UI, label it, and push the annotated dataset to the Hub. All this without a line of code! 𝗪𝗵𝘆 𝘀𝗵𝗼𝘂𝗹𝗱 𝘆𝗼𝘂 𝘂𝘀𝗲 𝗶𝘁? Your AI project's impact depends heavily on the effort and care you put into your data. This new feature enables you to iterate faster and make annotated data available in the right format for training and evaluation. 𝗛𝗼𝘄 𝗱𝗼𝗲𝘀 𝗶𝘁 𝘄𝗼𝗿𝗸? 1️⃣ Import initial data from a CSV or any format to Hugging Face 2️⃣ Load it into the Argilla UI and configure the annotation task 3️⃣ Annotate your dataset 🚀 Click on “Push to Hub” and share the dataset with your team (or the entire world) 👉 𝗥𝗲𝗮𝗱𝘆 𝘁𝗼 𝘁𝗿𝘆 𝗶𝘁 𝗼𝘂𝘁? Get started here: https://lnkd.in/dhA-swR5 Release highlights: https://lnkd.in/dbdQXG-W

Hot take: shipping BERT-sized models in 2025 will benefit far more people than sharing an LLM overfitted to some saturated leaderboards We're already seeing ModernBERT finetunes on the Hugging Face Hub. My guess is we'll see hundreds of these by the end of 2025.

🤯🔥LEARN HOW TO CREATE interactive agentic chatbots using Gradio that are capable of showcasing the Thoughts, Tasks, and interim responses of Multiple Agents as you await the final answer from your AI assistant. 🎯 Customer Support multi-agents with CrewAI and Gradio Showcasing here, a user-friendly, high-performing multi-agent gradio app. TO operate it, simply enter a webpage URL along with your questions related to that page, and in turn receive a high-quality response from the CrewAI Multi-Agent setup. 🚀Access this app on Hugging Face Spaces: https://lnkd.in/g6kXp_D2

In the past few months, we've invested a lot of efforts in improving the user management features of the Hugging Face hub that more than 5M AI builder are now using. It helps not only for easier organization collaboration but also for security (for example to make sure ex team members don't still have access to private models). If your manager, VP AI or admin/CISO is not aware, mention them below so that we can connect if they have any questions or feedback as most of these features are part of the Enterprise hub subscriptions: https://lnkd.in/e-RY-3vs) Cheers!