CPM.cu

Lightweight CUDA implementation for maximum LLM inference performance on edge GPUs (RTX series and Jetson).