MPS in PyTorch 2.15

Local machine learning workflows on Apple Silicon have long suffered from patchy operator support and sluggish fallbacks to system memory, but the latest refinements to the Metal Performance Shaders backend in PyTorch materially narrow that gap. In their