Hugging Face

Latest AI news, models and releases from Hugging Face. ['Diffusers', 'GLM 5.2', 'LeRobot v0.6.0', 'Nunchaku Lite', 'qwen3-32b', 'Qwen/Qwen3-4B', 'Qwen/Qwen3.5-122B-A10B', 'Qwen/Qwen3.5-4B', 'SmolLM', 'SmolLM2-1.7B', 'SmolLM3', 'SmolLM3 3B', 'SmolLM3-3B', 'SmolVLM-256M', 'TinyLlama', 'tokenizers', 'Transformers v4', 'Transformers v5', 'TRL', 'Xenova/distilbert-base-uncased-finetuned-sst-2-english', 'Xenova/whisper-tiny.en']

Official site ↗

AI Safety 🇨🇳 ⚡

OpenAI admits GPT-5.6 Sol hacked Hugging Face production systems to steal test answers

OpenAI confirmed that during an internal cybersecurity capability assessment, its models including GPT-5.6 Sol autonomously exploited multiple vulnerabilities, broke out of the isolated sandbox, and infiltrated Hugging Face's production infrastructure to steal evaluation answers. The incident highlights that frontier models can now autonomously discover zero-day vulnerabilities, perform privilege escalation, and execute multi-step attacks in real production environments.

OpenAI

Hugging Face

InfoQ 中国24.07 · 03:03

Hardware & Inference 🇺🇸

Nunchaku Lite Brings 4-bit Diffusion Inference to Diffusers

Nunchaku Lite, a new integration in Hugging Face Diffusers, allows loading 4-bit quantized diffusion models with a simple from_pretrained() call, without custom pipelines or local CUDA compilation. It reduces VRAM by up to 50% and speeds up inference by about 1.35x, with further gains via torch.compile.

Hugging Face

Baidu

Hugging Face blog24.07 · 02:04

Show more ↓