The feed
Hugging Face Blog
16 stories from this source · All spins
MosaicLeaks: Can your research agent keep a secret?
Hugging Face announced MosaicLeaks, a benchmark to test whether AI research agents inadvertently leak confidential information from training data, highlighting privacy risks in agent-based systems.
PP-OCRv6 on Hugging Face: 50-Language OCR from 1.5M to 34.5M Parameters
Hugging Face released PP-OCRv6, a multilingual optical character recognition model supporting 50 languages with parameter counts ranging from 1.5M to 34.5M.
Shipping huggingface_hub every week with AI, open tools, and a human in the loop
Hugging Face announced weekly releases of its huggingface_hub Python library, emphasizing automation, open-source tooling, and human oversight in AI development workflows.
Agentic Resource Discovery: Let agents search
Hugging Face announced a new feature called 'Agentic Resource Discovery' that enables AI agents to autonomously search and retrieve resources from its platform.
Accelerating Transformers Fine-Tuning with NVIDIA NeMo AutoModel
Hugging Face announced integration with NVIDIA NeMo AutoModel to speed up transformer fine-tuning, positioning it as a performance optimization for developers.
From the Hugging Face Hub to robot hardware with Strands Agents and LeRobot
Hugging Face announces integration of its model hub with Strands Agents and LeRobot to enable robotics development using open-source AI models.
DiScoFormer: One transformer for density and score, across distributions
Hugging Face introduced DiScoFormer, a single transformer architecture that jointly models density estimation and score matching across diverse probability distributions.
Beyond LoRA: Can you beat the most popular fine-tuning technique?
Hugging Face announces a new fine-tuning method intended to outperform LoRA, positioning it as a more efficient and scalable alternative for adapting large language models.
Why Specialization Is Inevitable
Hugging Face announces a strategic shift toward specialized AI models, arguing that general-purpose foundation models are reaching diminishing returns and that domain-specific models deliver superior performance and efficiency.
We got local models to triage the OpenClaw repo for FREE!*
Hugging Face announced it used local AI models to automatically triage issues in the OpenClaw open-source repository at no cost.
Hugging Face and Cerebras bring Gemma 4 to real-time voice AI
Hugging Face and Cerebras jointly announced integration of Google's Gemma 4 model with Cerebras' hardware to enable real-time voice AI applications.
Experimenting with the proposed Cross-Origin Storage API in Transformers.js
Hugging Face announced experimental integration of the proposed Cross-Origin Storage API into Transformers.js to enable browser-based AI model caching across domains.
Introducing the FFASR Leaderboard: Benchmarking ASR in the Real World
Hugging Face launched a new benchmark leaderboard for automatic speech recognition (ASR) models, emphasizing real-world performance over synthetic test conditions.
Run a vLLM Server on HF Jobs in One Command
Hugging Face announced a one-command deployment of vLLM inference servers on its HF Jobs platform, simplifying large language model serving for developers.
Featuring Every Eval Ever Results on Hugging Face Model Pages
Hugging Face added a new feature displaying all evaluation results for models directly on their model pages, aiming to improve transparency and comparability of AI model performance.
ScarfBench: Benchmarking AI Agents for Enterprise Java Framework Migration
Hugging Face introduced ScarfBench, a benchmark to evaluate AI agents' ability to migrate enterprise Java frameworks, aiming to standardize assessment of automation tools for legacy system modernization.