Four innocent lines of PyTorch hide a video preprocessing footgun.
In our H100 benchmark, the resulting physical plan made the model’s first operator run 8.9× slower.
The code is clean. The physical plan isn’t. 🧵
Spiral
@SpiralDB
- @mbakovic__ : If your data stack could survive an army of agents curating a data asset for investigations or experiments, this blog post is not for you. For everyone else, read the full post. spiraldb.com/blog/reversibi… 1/n
- Come meet researchers & engineers working on world models and multimodal AI Technical talks, curated attendee list, food & drinksWe're pumped to be hosting an event on Tuesday night (June 16) at the @krea_ai HQ in SF with our friends @TigrisData there will be deeply technical lightning talks from frontier AI labs like @metamorphiclabs, infra providers like @trychroma, and special guest! Details below 👇
