Models & Labs
NVIDIA Nemotron 3 Embed Tops RTEB Leaderboard
Hugging Face Bloghigh confidence
Why it matters
- →Nemotron 3 Embed models set a new standard in retrieval quality, topping the RTEB leaderboard.
- →The models offer open weights and datasets, allowing for extensive customization and deployment flexibility.
- →Optimized for high-throughput deployment, they enhance retrieval efficiency and reduce operational costs.
NVIDIA has released the Nemotron 3 Embed, a collection of embedding models that have achieved top rankings on the RTEB leaderboard. The 8B model leads in retrieval quality, while smaller variants offer efficient deployment options. These models are designed for production-scale retrieval, supporting multilingual and code retrieval tasks. With open weights and datasets, developers can customize and deploy these models on their infrastructure. The release marks a significant advancement in retrieval technology, providing enhanced tools for developers to improve retrieval efficiency and accuracy.
Read originalMore from Hugging Face Blog
© Hugging Face BlogModels & Labsmodels
DharmaOCR Outperforms Newer OCR Models
DharmaOCR has demonstrated a significant advantage over newer OCR models like Mistral OCR4 and Unlimited-OCR in handling Brazilian Portuguese documents. This success is attributed to its domain-specific training, which focuses entirely on the linguistic nuances of Brazilian Portuguese. By employing a two-stage training process, including Direct Preference Optimization, DharmaOCR achieves higher extraction quality and stability. This specialization allows it to outperform more generalized models, highlighting the benefits of targeted training over broader multilingual approaches.
Hugging Face Blog
© Hugging Face BlogMarket & Regulationother
Hugging Face Discloses Security Breach
Hugging Face has reported a security breach involving unauthorized access to internal datasets and credentials. The attackers exploited vulnerabilities in the data-processing pipeline, gaining escalated access within internal clusters. AI-assisted anomaly detection played a crucial role in identifying the breach, showcasing the essential role of AI in modern cybersecurity. In response, Hugging Face has patched the vulnerabilities, rotated credentials, and strengthened security protocols. This incident highlights the increasing sophistication of AI-driven attacks and the necessity for advanced defensive measures. As autonomous attack tools become more common, having robust AI-driven defenses is critical for online platforms.
Hugging Face Blog
© Hugging Face BlogAgentsagents
Building Reliable AI Agents for Maritime Awareness
Hugging Face's Shippy project tackles the complexities of developing AI agents for critical applications like maritime domain awareness. The project focuses on ensuring accuracy and trustworthiness by employing a layered approach, including a deterministic CLI and structured agent skills. Shippy is designed to operate within strict boundaries, avoiding speculation and legal judgments, which ensures it delivers precise information to maritime analysts. This design allows Shippy to manage complex queries effectively, providing dependable insights while maintaining data security and session isolation. By integrating AI into real-world workflows, Shippy demonstrates how AI can be both precise and secure in high-stakes environments.
Hugging Face Blog
More in Models & Labs
Models & Labsmodels
llama.cpp b10043 release supports CUDA Virtual Devices
The b10043 release of llama.cpp marks a notable enhancement with the addition of CUDA Virtual Devices, which significantly improves GPU resource management. By removing the NCCL path when virtual devices are in use, the update fine-tunes performance for these specific setups. This release also includes a comprehensive code refactor and the implementation of GPUx2 server CI jobs, reflecting a commitment to better testing and deployment processes. While there are no new model architectures, the update enhances the platform's flexibility across various operating systems, making it more adaptable for developers working with a wide range of hardware configurations.
llama.cpp Releases
Models & Labsmodels
Llama.cpp b10051 Release Enhances SME Kernel Dispatch
The latest release of llama.cpp, b10051, addresses a critical issue in kernel dispatch by distinguishing between SME and SME2 capabilities. Previously, the integration treated SME as a single capability, leading to incorrect dispatch on SME(v1)-only hardware due to the use of SME2-specific instructions. This update introduces both build-time and runtime distinctions, ensuring that kernels are dispatched based on actual hardware support. This refinement enhances the accuracy and efficiency of operations on different hardware configurations, marking a significant improvement for developers working with these systems.
llama.cpp Releases
© The Verge AIModels & Labsproductivity
Google renames NotebookLM to Gemini Notebook
Google's AI note-taking app is getting a new identity as it transitions from NotebookLM to Gemini Notebook. This rebranding aligns with deeper integration into Google's broader AI ecosystem, including Gemini and Google Search. The app, initially introduced as Project Tailwind, has evolved with features like AI-generated podcasts and TikTok-style clips. Now, it also offers a secure cloud computing feature for code execution, initially available to select business customers. This move signifies Google's commitment to enhancing AI-driven productivity tools.
The Verge AI