PyTorch migrates MPS abs tests to OpInfo for better coverage

PyTorch has merged pull request #190435, which migrates large absolute value (abs) test samples from the MPS-specific test file (test_mps) into the centralized OpInfo test framework. This is part of an ongoing effort to consolidate operation tests across all hardware backends, ensuring that critical operations like abs are validated uniformly whether running on CPU, CUDA, or MPS (Apple Silicon). The migration follows a review of a previous PR (#190053) and was co-authored by Claude Opus 4.8, an AI model from Anthropic, highlighting the growing trend of AI-assisted code contributions in open-source projects.

The OpInfo framework provides a structured way to define operation metadata, including sample inputs, reference implementations, and error tolerances for different dtypes and devices. By moving MPS-specific abs tests into OpInfo, PyTorch makes it easier to add new test cases, catch regressions, and maintain consistency across backends. For developers building on Apple Silicon Macs, this means more reliable MPS performance for deep learning workloads that rely on absolute value computations, such as loss functions and optimization algorithms. The change also reduces code duplication and makes the test suite more scalable as PyTorch continues to expand support for diverse hardware.