Falcon-Emirati: When an LLM Learns the Dialect, the Culture, and the Nuance
AI summary Falcon-Emirati-7B is a language model designed to understand and generate Emirati Arabic, focusing on dialect, culture, and nuance.
News & blogs · AI & ML
Hugging Face Blog
The latest from Hugging Face Blog, part of our AI & ML coverage. Updated automatically as new posts are published.
Open-sourcing AstaBrief, the fast report-generation model in Asta
A Blog post by Ai2 on Hugging Face
AutoSynthData: Generating Training Data for Enterprise Agents
AI summary AutoSynthData generates training data for enterprise agents by identifying weaknesses and creating new tasks to improve capabilities.
Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEs
A Blog post by Ai2 on Hugging Face
Open TTS Leaderboard: Scalable Evaluation for Multilingual Text-to-Speech and Voice Cloning
AI summary The Open TTS Leaderboard evaluates text-to-speech models using metrics such as WER and speaker similarity, and allows users to compare and provide feedback on model outputs.
NVIDIA Kumo Tabular Sets a New Accuracy-Efficiency Frontier for Tabular Prediction
AI summary NVIDIA Kumo Tabular is an open foundation model for tabular data that predicts labels of new rows in a single forward pass without training or tuning.
Getting the Source Right, Not Just the Fact: Source-Aware Verification for MCP Agents
A Blog post by Multiverse Computing on Hugging Face
Holo4: powering generalist computer-use agents
A Blog post by H company on Hugging Face
Welcome RL Environments to the hub
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Accelerating vision-language models with LFM2.5-VL-DSpark
A Blog post by Liquid AI on Hugging Face
How UK AISI and EvalEval Are Making Benchmark Results Reproducible
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Transformers now runs llama.cpp quants
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Jun Kim, oMLX creator and maintainer, joins Hugging Face to support the MLX community
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
tokenizers v1: encode, decode and scaling, measured
AI summary Hugging Face has released a new version of the tokenizers library, v1, which improves performance and scalability for machine learning workflows.
Your Agent Aced the Task. Will It Do It Again?
AI summary A new metric, Pass^k, measures an agent's consistency by checking if it succeeds on all k runs of a task, addressing a gap in standard evaluation metrics.
Rebuilding AUTOMATIC1111 with Gradio Workflow
AI summary Hugging Face has rebuilt AUTOMATIC1111's feature set as a single workflow canvas called Workflow1111, using 73 nodes and 11 media pipelines.
NeoMME: an efficient Multimodal-native and Multilingual Encoder
AI summary NeoMME is a multilingual, multimodal encoder that generates vector representations for text and images using a single Transformer encoder, trained from scratch with a masked…
Give Your Coding Agents a Memory You Own
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Training a coding model to paint watercolours with TRL and OpenEnv
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
BenchMIRT: What are LLM benchmarks actually measuring?
A Blog post by Ai2 on Hugging Face
The Open ASR Leaderboard Adds Its First Global South Language
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers
AI summary The Hugging Face library has introduced a new model type, MultiVectorEncoder, for ColBERT-style late interaction retrieval, and provides a training approach for it.
Granite 4.2 LLMs: How They're Built
AI summary The Granite 4.2 LLM family is a reasoning-focused release with three sizes: 3B, 8B, and 30B. They are post-trained from Granite-4.1 base models.
Quantization-Aware Healing: a compressed, 4-bit model that outperforms its full-precision original
A Blog post by Multiverse Computing on Hugging Face
Wire It, Run It, Deploy It: AI Workflows in Gradio
AI summary Gradio's gr.Workflow allows users to create and deploy AI pipelines as graphs of nodes, enabling parallel processing and real-time results.
How Hugging Face Inference Endpoints, Jobs, and Buckets Power Search on Papers with Code
AI summary Hugging Face built a hybrid search system for Papers with Code, combining keyword and vector search, and learned several lessons from its development and deployment.