Add explicit support for optimized GPU-to-GPU data transfers in multi-GPU environments. ## TODO - [ ] Detect GPU topology and P2P capability - [ ] Enable peer access when available - [ ] Use explicit P2P copy for GPU↔GPU transfers - [ ] Avoid cross-GPU intermediate tensors by default - [ ] Provide basic logging for transfer paths
Add explicit support for optimized GPU-to-GPU data transfers in multi-GPU environments.
TODO