Skip to content
Live
Trending storyPast story

Hugging Face introduces PyTorch 2.15 MPS backend update

1 reports1 outlets2026-10-20Update

The story so far

AI overview

The PyTorch team at Hugging Face published a blog post introducing updates to PyTorch 2.15 on the Apple MPS backend: the new torch.float8_e4m3fn tensor and F.scaled_mm, with FP8 tensors occupying half the space of BF16, and claims of faster LLM decoding and Metal kernel improvements. According to the blog, MPS now requires macOS 15 or higher, and PyTorch 2.15 requires Python 3.11 or higher; candidate versions for trial should be done in isolated environments. The blog also mentions that logspace does not currently support integer or boolean outputs with more than one element. All the effects mentioned are claims from the team in the introduction.

AI-written from the reports · Updated 27 days ago

Coverage timeline

Follow the coverage to see every side of the story.

20 Oct
  1. Hugging Face BlogFeatured
    MPS in PyTorch 2.15

    The Hugging Face PyTorch team introduces updates to the MPS backend in PyTorch 2.15, adding the tensor torch.float8_e4m3fn and F.scaled_mm, with FP8 tensors occupying half the space of BF16.