LM/llama.cpp for CUDA

Build

cmake -B build/cuda -G Ninja \
  -DGGML_CUDA=ON \
  -DCMAKE_BUILD_TYPE=Release

cmake -B build/cuda -G Ninja \
  -DGGML_CUDA=ON \
  -DGGML_NATIVE=ON \
  -DGGML_LTO=ON \
  -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_INSTALL_RPATH="\$ORIGIN/../lib" \
  -DCMAKE_CXX_FLAGS="-O3 -fp-model=fast=2 -qopenmp" \
  -DCMAKE_C_FLAGS="-O3 -fp-model=fast=2 -qopenmp"

cmake --build build/cuda
cmake --install build/cuda --prefix ~/llama.cpp/cuda