Pipeline Parallelism is here: 8x faster than Apple's WWDC26 fine-tuning

Re:Pipeline Parallelism is here: 8x faster than Apple's WWDC... by klay_moon on Wednesday November 11, 2026 @03:03PM (#50457912)

Hi, I am Klay, founder of NukeTorch. We are a small team who is proud to beat Apple by 8x using the exact same recipe as WWDC26 Qwen 9B LoRA fine-tuning with our own Tensor Engine. We hope you will test-out NukeTorch for a single Mac from our Github/Hugging Face. If you have four or more of the same Macs, we want you to test our Pipeline Parallelism for Training and Inferencing.

https://developer.apple.com/videos/play/wwdc2026/233/?time=1...

https://github.com/NukeTorch/nuketorch https://huggingface.co/NukeTorch https://x.com/NukeTorchAI [email protected]

[ Parent | on HN ]

There are no comments on this comment yet.