Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox
AI summary ThinkingBox-Training uses database state checks as rewards for post-training agents, improving success rates.
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.
OlmoEarth v1.3: State-of-the-art embeddings for the whole planet
AI summary Hugging Face released OlmoEarth v1.3, a satellite embedding model with precomputed global data and a browser tool for exploration.
Ensemble forecasting with IBM Granite time-series models
AI summary IBM Granite time-series models use an ensemble approach for forecasting, combining four architectures for better accuracy and flexibility.
NVIDIA Kumo Relational Outperforms Supervised Machine Learning on Relational Data, No Training Required
A Blog post by NVIDIA on Hugging Face
How Specializing the Vocabulary of SLMs Can Cut Inference Cost by 77% in Real Applications
AI summary Specializing model vocabulary can reduce inference costs by 77% without changing the base model, as shown in a production essay grader.
Profiling TorchNeuron (Part 1): System-level profiling on AWS Trainium.
AI summary This article explains system-level profiling on AWS Trainium using TorchNeuron, focusing on analyzing inference performance and using tools like Neuron Explorer.
Profiling TorchNeuron (Part 1): System-level profiling on AWS Trainium.
AI summary This article explains system-level profiling on AWS Trainium using TorchNeuron, focusing on analyzing inference performance and using tools like Neuron Explorer.
Granite Speech 5.0: Transcribe, Translate, and Attribute more Accurately and Faster
AI summary Hugging Face announces updates to IBM's Granite Speech collection, including faster and more accurate models for transcription and translation.
NVIDIA Kumo Forecast Combines Pretrained Time-Series Representations with Historical Pattern Retrieval and Interpretability
AI summary NVIDIA Kumo Forecast combines a pretrained Transformer with historical data retrieval for time-series forecasting and interpretability.
Announcing SimReady Central: An Open Library of Simulation-Ready 3D Assets for Robotics and AI Factories
AI summary Hugging Face launches SimReady Central, a library of 2,000 simulation-ready 3D assets for robotics and AI factories.
NVIDIA Cosmos3-H-Dreams: Bringing Clinical Surgery to Life Through Generative Simulation
A Blog post by NVIDIA on Hugging Face
FairStress: A deeper evaluation of fairness in LLM decision-making
A Blog post by Ai2 on Hugging Face
ALTK-Evolve: How learned memory tops the τ²-bench leaderboard on customer service tasks without a frontier model
A Blog post by IBM Research on Hugging Face
Hugging Face ❤️ TorchTPU: TPUs, Natively
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
AnyLanguageModel 1.0: One Year of Local Models on Apple Platforms
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
MPS in PyTorch 2.15
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
TorchNeuron: running HF models on AWS Trainium is easier than ever
AI summary TorchNeuron provides a new PyTorch backend for AWS Trainium, making it easier to run and share models on the Hugging Face hub.
Scaling Indic Pretraining: 2.8 Trillion Tokens and an Open Data Generation Pipeline
AI summary Hugging Face introduces a 2.8 trillion token pretraining corpus for 22 Indian languages, along with an open data pipeline.
Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint
AI summary Granite Switch 4.1 allows efficient switching between chat, decisions, and speech tasks from one endpoint with fast performance.
Apertus 1.5 Release Notes
A Blog post by Swiss AI Initiative on Hugging Face
Welcome Muse Spark Open: open-weight frontier multimodal mixture-of-experts model from Meta
AI summary Muse Spark Open is a large multimodal model from Meta with 816 billion parameters, supporting text, images, audio, and video inputs.
Welcome FLUX.3 Image Klein in Diffusers
AI summary FLUX.3 Image Klein is now available in Diffusers, offering text-to-image, image-to-image, and layout control features.
HiRO-ACEv2: Simulating extreme weather at 3 km resolution for a fraction of the cost
AI summary HiRO-ACEv2 simulates extreme weather at 3 km resolution with high accuracy and lower cost than existing models.
Run fast, private agents on-device with LFM2.5-24B-A2B
AI summary LFM2.5-24B-A2B is a 24B-parameter model with 2B active parameters, designed for fast, private on-device agents.
Impactful scheduling for GPU clusters
AI summary Ai2 improved GPU scheduling by replacing a priority system with a budget-based approach, increasing transparency and fairness in resource allocation.
The model that didn't exist, so you made it yourself
AI summary A user created several models using ML Intern, resulting in smaller, efficient versions of existing models.
Multimodal open d1 decision models for the edge
AI summary Hugging Face released two open decision models, d1-3B and d1-omni-600M, built on Liquid Foundation Models for edge and GPU inference.
Introducing Falcon ASR
A Blog post by Technology Innovation Institute on Hugging Face
One Model Family, Two Gold-Level Results: Fine-Tuning Nemotron for IOI and IMO
AI summary Nemotron was fine-tuned to achieve gold-level results in IOI and IMO competitions, showing its adaptability.