Optimize for GPU skill: what it does and how to install it

What it does. Profiles a CPU baseline, selects the least disruptive CUDA or RAPIDS path, moves data through a coherent GPU pipeline, and compares outputs with numerical tolerances. It benchmarks synchronously, including transfers and memory, then keeps, revises or rejects the port.

When to use it. For large scientific Python workloads involving arrays, dataframes, machine learning, graphs, images, simulations, vector search or GPU file I/O. It keeps a CPU path for small, sequential or unsupported workloads and does not port code already well served by a GPU-native framework.