Hugging Face Blog
AI news source. Articles are auto-selected and adapted by Hamidun News editors.
Latest publications
IBM released Granite Embedding R2 — a multilingual model for semantic search
IBM introduced Granite Embedding R2, an open multilingual model for semantic search with 32K context support and best-in-class performance among sub-100M models.
H Company released Holotron-12B — a model for agents with a 2x speed increase
H Company published Holotron-12B on Hugging Face: the multimodal model for AI agents delivers more than a 2x throughput gain in interface-use tasks on a single H100.
NVIDIA introduced SPEED-Bench — a unified benchmark for speculative decoding
NVIDIA published SPEED-Bench, a dataset and measurement framework that compares speculative decoding across real-world workloads, long contexts, and different inference engines.
IBM released Mellea 0.4.0 and Granite Libraries for verifiable AI pipelines
IBM Research updated the open-source Mellea framework to version 0.4.0 and released three Granite Libraries for structured, verifiable, and safe AI workflows.
NVIDIA showed how to fine-tune an embedding model for a specific domain in a day
NVIDIA and Hugging Face published a step-by-step recipe that turns a base embedding model into specialized search over internal documents in a few hours.
ServiceNow introduced EVA — a new framework for evaluating voice AI agents
ServiceNow released EVA — a system that evaluates voice AI agents not only by task success, but also by dialogue quality, from response brevity to turn timing.
IBM releases Granite 4.0 3B Vision for extracting data from documents and charts
IBM introduced Granite 4.0 3B Vision, a compact multimodal model for extracting tables, charts, and key fields from documents that can be integrated into enterprise pipelines with Docling.
H Company introduces Holo3 — an AI agent for computer use with a record score on OSWorld-Verified
H Company has released Holo3, a model for computer use that scored 78.85% on OSWorld-Verified and was trained on synthetic enterprise scenarios.
Google released Gemma 4 on Hugging Face: multimodal models for local inference
Google DeepMind has opened the Gemma 4 family on Hugging Face: four multimodal models under the Apache 2.0 license, with up to 256K context and deployment ranging from phones to workstations.
Hugging Face added gradio.Server: custom frontends can now connect to a Gradio backend
Hugging Face’s new gradio.Server turns Gradio into a backend layer for React, Svelte, and plain HTML/JS, while preserving request queues, ZeroGPU, and compatibility with Spaces.
Hugging Face transfers Safetensors to the PyTorch Foundation for neutral governance of the format
Hugging Face announced that Safetensors has become a PyTorch Foundation project: there are no breaking changes for users, while the format's development moves to a neutral governance model.
Overworld released Waypoint-1.5: 720p interactive worlds for consumer GPUs
Overworld released Waypoint-1.5, a world model for running locally on consumer GPUs: up to 720p and 60 FPS, plus a lighter 360p version for a wider range of PCs and laptops.
Hugging Face released a Skill for quickly porting Transformers models to MLX
Hugging Face introduced a Skill and a separate test harness to port new models from Transformers to mlx-lm on MLX almost immediately, without a stream of raw AI-generated PRs.
IBM Research analyzed where AI agents break down on APIs, documents, and rules in VAKRA
IBM Research's analysis of VAKRA shows that even strong models lose reliability when they have to combine APIs, documents, multi-step reasoning, and tool constraints.
Hugging Face published Ecom-RLVE, a training environment for e-commerce AI agents
Hugging Face published Ecom-RLVE, an open-source environment where AI agents learn to handle purchase conversations, use tools, and earn a verifiable reward for the cart they actually assemble.
QIMMA Arabic LLM Leaderboard: Qwen3.5 Leads, Karnak Second
TII launched QIMMA, an Arabic LLM leaderboard that first checks the benchmarks themselves and only then compares models on 52,000 examples from seven domains.
NVIDIA introduced Nemotron OCR v2: multilingual OCR trained on 12.2 million synthetic documents
NVIDIA showed how it built Nemotron OCR v2: the model was trained on 12.2 million synthetic documents to recognize multiple languages with a single engine and process up to 34.7 pages per second.
NVIDIA showed how Gemma 4 with voice and a webcam runs on Jetson Orin Nano Super
NVIDIA published a demo in which Gemma 4 decides on its own when to activate the webcam and responds by voice — all locally on Jetson Orin Nano Super with 8 GB of memory.
NVIDIA introduces NeMo Retriever — agentic search for complex enterprise data
NVIDIA showcased an agentic pipeline in NeMo Retriever: the system goes beyond semantic search, planning steps, refining queries, and has already taken first place in ViDoRe v3.
Nvidia unveiled the first open dataset and foundation AI models for medical robots
Nvidia and partners on Hugging Face released the first large open dataset for medical robots and two foundation models for surgery, simulation, and future autonomy.
NVIDIA released Nemotron 3 Nano 4B — a compact hybrid model for on-device deployment
NVIDIA made the 4B Nemotron 3 Nano model with a hybrid Mamba-Transformer architecture available: the lowest VRAM usage in its class, 18 tokens/s on Jetson Orin Nano, and open weights.
Hugging Face: Chinese open-source models overtake the US in AI ecosystem downloads
Hugging Face showed that open-source AI nearly doubled in scale over the past year, while Chinese models already account for 41% of downloads and set the pace in releases, adaptation, and local deployment.
AI Model Evaluation Now Costs More Than Training — A New Barrier for Researchers
EvalEval Coalition analyzed the cost of AI-benchmarks: a single agentic test costs $40,000 or more, and academic groups can no longer afford independent evaluation.
IBM reveals how it built Granite 4.1: 15 trillion tokens, 512K context window, and focus on quality
IBM detailed its approach to training Granite 4.1: five pretraining stages, 15 trillion tokens, context window up to 512K, and separate SFT and RL pipelines for quality improvement.
Hugging Face adds DeepInfra to Inference Providers for unified model API
Hugging Face connected DeepInfra to Inference Providers: DeepSeek, Kimi, and GLM models can now be run from Hub pages, via SDK, and through the unified router without separate integration.
NVIDIA Introduced Nemotron 3 Nano Omni for Long Documents, Audio, Video, and AI Agents
NVIDIA introduced Nemotron 3 Nano Omni — an open multimodal model for long documents, audio, video, and GUI scenarios with emphasis on speed and context.
Hugging Face Explains Fine-tuning of Multimodal Embeddings and Reranker Models
Hugging Face released a practical guide on training multimodal embedding and reranker models in Sentence Transformers and demonstrated how domain-specific fine-tuning improves document retrieval.
How Hugging Face Builds Scalable Web Apps with OpenAI Privacy Filter
Hugging Face demonstrated three scenarios for OpenAI Privacy Filter: document reading with PII highlighting, image anonymization, and secure pastebin with public and private versions.
Hugging Face: open-source AI gives defenders the same capabilities as attackers
Hugging Face explains why open models and tools are a structural advantage in cybersecurity, not a threat.
Hugging Face trained an image generation model in 24 hours
The third part of Hugging Face's PRX project shows that a full-fledged text-to-image model can be trained in just 24 hours. This changes perceptions of the accessibility of generative AI.