Learn to write performant kernels in Metal Shader Langauge

Breaking down how torch.compile() optimizes large models