Squeeze-and-Excitation Networks

[ARCHIVED] Cooperative primitives for CUDA C++. See https://github.com/NVIDIA/cccl

An efficient GPU support for LLM inference with x-bit quantization (e.g. FP6,FP5).