Multimodal open d1 decision models for the edge Opens at the publisher in a new tab

No excerpt provided for multimodal edge decision models.

Introducing Falcon ASR Opens at the publisher in a new tab

Falcon ASR introduced.

One Model Family, Two Gold-Level Results: Fine-Tuning Nemotron for IOI and IMO Opens at the publisher in a new tab

Nvidia demonstrates fine-tuned Nemotron model achieving gold-level results on math benchmarks.

Falcon-Emirati: When an LLM Learns the Dialect, the Culture, and the Nuance Opens at the publisher in a new tab

Language model trained on Emirati dialect to capture cultural and linguistic nuance.

The Agent Said It Was Done. The Database Disagreed. Opens at the publisher in a new tab

AI agent reported task completion but database records showed it had failed to execute properly.

Open-sourcing AstaBrief, the fast report-generation model in Asta Opens at the publisher in a new tab

Asta open-sources AstaBrief, a fast report-generation model.

AutoSynthData: Generating Training Data for Enterprise Agents Opens at the publisher in a new tab

System for automatically generating synthetic training data to teach enterprise AI agents.

Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEs Opens at the publisher in a new tab

Allen Institute releases open training infrastructure for large mixture-of-experts AI models.

Open TTS Leaderboard: Scalable Evaluation for Multilingual Text-to-Speech and Voice Cloning Opens at the publisher in a new tab

Open TTS Leaderboard provides scalable evaluation metrics for multilingual text-to-speech and voice cloning systems.

NVIDIA Kumo Tabular Sets a New Accuracy-Efficiency Frontier for Tabular Prediction Opens at the publisher in a new tab

NVIDIA Kumo Tabular achieves new accuracy-efficiency benchmark for tabular prediction.

Getting the Source Right, Not Just the Fact: Source-Aware Verification for MCP Agents Opens at the publisher in a new tab

Research explores source verification methods for AI agents to improve reliability and accuracy.

Holo4: powering generalist computer-use agents Opens at the publisher in a new tab

Holo4 enables generalist computer-use agents to perform multiple tasks across applications.

Welcome RL Environments to the hub Opens at the publisher in a new tab

Accelerating vision-language models with LFM2.5-VL-DSpark Opens at the publisher in a new tab

LFM2.5-VL-DSpark accelerates vision-language models for improved performance on multimodal tasks.

How to Use NVIDIA Warp and MjWarp to Accelerate Robotics Simulation and Learning Workflows Opens at the publisher in a new tab

NVIDIA releases Warp and MjWarp tools to accelerate robotics simulation and learning.

**Know Who Spoke When: Build Real-Time, Multi-Speaker AI with NVIDIA Nemotron 3 Diarization** Opens at the publisher in a new tab

NVIDIA Nemotron 3 enables real-time multi-speaker identification in AI voice applications.

How UK AISI and EvalEval Are Making Benchmark Results Reproducible Opens at the publisher in a new tab

UK AISI and EvalEval work to make AI benchmark results reproducible and verifiable.

Transformers now runs llama.cpp quants Opens at the publisher in a new tab

Technical software update enabling Transformers library compatibility with llama.cpp quantized models.

Jun Kim, oMLX creator and maintainer, joins Hugging Face to support the MLX community Opens at the publisher in a new tab

MLX creator Jun Kim joins Hugging Face to support the open source machine learning framework community.

Pruning LLMs Like a Physicist: Block Removal as an Ising Optimization Problem Opens at the publisher in a new tab

Researchers propose block removal method using physics-based optimization to prune large language models efficiently.

tokenizers v1: encode, decode and scaling, measured Opens at the publisher in a new tab

Analysis of tokenizer performance covering encoding, decoding and scaling characteristics with measurements.

Your Agent Aced the Task. Will It Do It Again? Opens at the publisher in a new tab

Article addresses consistency and reliability of AI agent performance across repeated tasks.

Async GRPO with LoRA across HF Jobs: a bucket, a proxy, and no NCCL Opens at the publisher in a new tab

Rebuilding AUTOMATIC1111 with Gradio Workflow Opens at the publisher in a new tab

Technical post on rebuilding image generation tool with Gradio workflow.