(gimletlabs.ai)

186 points nserrino | 1 comments | 03 Sep 25 17:03 UTC | HN request time: 0s | source

Show context

> non 100% correctness of kernels

wouldn't model not work properly if kernels are even slightly off?

wasn't kernels a part of training stack for models? am I missing anything?

1. arjvik ◴[03 Sep 25 22:15 UTC] No.45120966[source]▶

I believe their speedup is computed _assuming they can easily fix the correctness bugs in the kernels_.

In practice, with slight differences the model will feel almost lobotomized.

Speeding up PyTorch inference on Apple devices with AI-generated Metal kernels