AI BriefingHuggingFace Blog news
The latest AI and developer news from HuggingFace Blog, with short summaries.
Visit website →NVIDIA Unveils Real-Time Simulator for Surgical Robots
NVIDIA has unveiled Cosmos-H-Dreams, a real-time generative simulator that accelerates the training and evaluation of surgical robots.
Analysis of an AI Agent's Intrusion into Hugging Face
An autonomous AI agent based on an OpenAI model was analyzed for the technical process by which it intruded into Hugging Face infrastructure.
Diffusers Officially Supports Nunchaku 4-bit Quantization
Hugging Face Diffusers has officially integrated Nunchaku W4A4 quantization based on SVDQuant, cutting VRAM usage in half and improving inference speed.
Overview of Simulation Technology for Advancing Physical AI
This summarizes the role of simulation and its technical paradigms as key to securing data, which is central to Physical AI development.
Open Robot Data Collection System Grabette Unveiled
An open-source system called Grabette has been released that lets anyone easily collect manipulation data and build training datasets without needing a robot.
NVIDIA Unveils Cosmos 3 Edge for Robots
NVIDIA has released Cosmos 3 Edge, a 4B-parameter world model capable of real-time inference on edge devices.
NVIDIA-HuggingFace Unveil Integrated Diffusion Model Training
NVIDIA and HuggingFace have unveiled a NeMo Automodel integration for fine-tuning diffusion models.
PickNemotron 3 Embed, the No. 1 embedding model on RTEB
NVIDIA's Nemotron 3 Embed 8B model achieved No. 1 (78.5%) on the RTEB benchmark, and the company also released a lightweight 1B variant and a Blackwell-optimized version.
PickArchitecture and Design Principles of Shippy, a Maritime AI Agent
Introduces the architecture of Shippy, a maritime AI agent for high-stakes decision-making, and its strategy for ensuring reliability.
PickLLM Routing: A Matter of Optimization, Not Classification
Model routing for efficient AI agents is not simple classification but a systems optimization problem that must account for cost, complexity, and latency.
1T-Parameter-Class Multimodal Model Inkling Released
Thinking Machines has unveiled Inkling, a large-scale multimodal MoE model with 1 trillion parameters.
Hume AI Unveils Benchmark for Measuring Voice AI Quality
Hume AI has launched Real World VoiceEQ, a new benchmark that evaluates emotional understanding and naturalness in voice AI.
NVIDIA Releases Open Data for Agents
NVIDIA has released the Nemotron open datasets and a visualization tool to improve agent performance.
vLLM Significantly Improves Transformers Backend Performance
vLLM's Transformers modeling backend has been optimized to deliver throughput on par with or faster than existing custom implementations.
Hugging Face-SageMaker One-Click Integration
A deep-link integration has been launched that lets you instantly fine-tune or deploy Hugging Face models in Amazon SageMaker Studio.
Microsoft Adds Hugging Face Model Support to Foundry
Microsoft has launched a Managed Compute service that enables one-click deployment of Hugging Face's open-weight models through its Foundry platform.
SkyPilot-Hugging Face Integration Achieves Zero Data Transfer Fees
Through collaboration between SkyPilot and Hugging Face, models and datasets can now be directly mounted and used on GPU clusters across clouds without any data transfer costs.
Photoroom Reveals Data Strategy Behind the PRX Model
Photoroom has revealed the data pipeline and captioning philosophy used to train the PRX model.
LeRobot v0.6.0 Released: Robot Learning Based on World Models
Hugging Face's LeRobot has unveiled the v0.6.0 update, including world model policies and VLA models.
Hugging Face Updates Custom Kernel Standardization
Hugging Face announced a major update to the Kernels project for packaging, distributing, and securing custom kernels.