trunk/07ff1a7dd9556373b5b01ee5bb5064bf07b1eb15: [local_tensor] Add standalone rank_map/tensor_map functions (#174795)
The new standalone functions work seamlessly across both local and real distributed tensor modes.
Deep Dive
Meta's PyTorch team (with contributions from Claude AI) has merged PR #174795, adding top-level rank_map and tensor_map functions. These functions work identically under LocalTensorMode (returning LocalTensor) and in real distributed environments (returning plain Tensor). Developers can now write distributed computing code that runs the same way locally and at scale, simplifying development workflows and testing for large-scale machine learning models.
Why It Matters
Simplifies distributed ML development by removing mode-specific code, making local testing more accurate.