HuggingFace
18 mentions across all digests
Hugging Face is an AI platform and model hub that publishes technical deep-dives on model architectures, hosts open-weight models, and develops tools including the transformers library used to implement and deploy large language models.
AI evals are becoming the new compute bottleneck
Evaluation benchmarks now cost $40K–$2.8K per run, making frontier-model testing prohibitively expensive and gatekeeping reproducible research—a shift where compute constraints moved from training to evaluation infrastructure.
Introducing NVIDIA Nemotron 3 Nano Omni: Long-Context Multimodal Intelligence for Documents, Audio and Video Agents
NVIDIA open-sources Nemotron 3 Nano Omni, a long-context multimodal model delivering 9x higher throughput than competitors while excelling at document, audio, and video understanding.
From Rainforests to Recycling Plants: 5 Ways NVIDIA AI Is Protecting the Planet
NVIDIA open-sources Earth-2, a climate forecasting AI model that accelerates weather predictions from hours to minutes, while allied applications achieve 90% waste diversion in recycling facilities.
QIMMA قِمّة ⛰: A Quality-First Arabic LLM Leaderboard
QIMMA reveals systematic quality issues in widely-used Arabic benchmarks, then consolidates 52K+ validated samples to build a quality-first leaderboard for Arabic LLMs.
We got 207 tok/s with Qwen3.5-27B on an RTX 3090
Hand-written CUDA kernels and speculative decoding achieve 207 tok/s for Qwen3.5-27B on consumer RTX 3090, proving open-source optimization can match commercial inference systems on commodity hardware.