Heidelberg AI News

Official announcement · Hugging Face ·

Transformers Model Integration Contracts

On July 30, 2026, Transformers enabled PyTorch's scaled dot-product attention (SDPA) for T5. As a side effect, UMT5's decoder could attend to future tokens. During training, it could peek at the answer it was supposed to predict. Nothing crashed, and the bug stayed in main for almost two months.

Read at huggingface.co The opening of the article at huggingface.co.

More from Hugging Face

All stories

OfficialHugging Face

nemotron-halloween

TL;DR: MonsterHouse is an interactive projection I created for Halloween. He listens to trick-or-treaters, answers in the voice of a grumpy old pirate ship and softens when visitors remember their manners.

OfficialHugging Face

Granite Switch 4.1.1 — Update models like you update software

Granite Switch 4.1.1 offers modular Granite models with built-in adapters that switch on as needed from a single checkpoint, delivering chat, decisions, and speech from one endpoint. The update also introduces Granite Speech 5.0 for more accurate and faster transcription, translation, and attribution.