Bitwise-identical training and inference across GPU vendors – Nvidia, AMD, Apple

The project delivers bitwise‑identical training and inference across Nvidia, AMD, and Apple GPUs, simplifying model portability between platforms. Adoption invo