Hugging Face has published a blog post titled "MPS in PyTorch 2.15," focusing on the Metal Performance Shaders backend for Apple GPUs. The post is aimed at PyTorch users who want to leverage Apple silicon for model training and inference.
The blog likely explains how to enable MPS, what operations are supported, and any performance considerations introduced in the 2.15 release. As a single source, it does not offer contrasting viewpoints, but it provides a practical reference for developers working with PyTorch on macOS.
For those following PyTorch's multi-backend strategy, this post is a useful update on the state of Apple GPU support. It does not cover other backends or compare MPS with CUDA, so readers should look elsewhere for broader benchmarks.