编译llama.cpp

# 可能会报版本过低：sudo apt install cmake
# 升级：pip install --upgrade cmake
pip install cmake
git clone https://github.com/ggerganov/llama.cpp --depth=1
cd llama.cpp/

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc source ~/.bashrc

WARNING: The scripts ccmake, cmake, cpack and ctest are installed in '/home/hzc/.local/bin' which is not on PATH.
Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location.

ls -d /usr/local/cuda*
#> /usr/local/cuda  /usr/local/cuda-12

which nvcc || find /usr/local -name nvcc 2>/dev/null
#> /usr/bin/nvcc

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-ubuntu2004.pin
sudo mv cuda-ubuntu2004.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget https://developer.download.nvidia.com/compute/cuda/12.4.0/local_installers/cuda-repo-ubuntu2004-12-4-local_12.4.0-550.54.14-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu2004-12-4-local_12.4.0-550.54.14-1_amd64.deb
sudo cp /var/cuda-repo-ubuntu2004-12-4-local/cuda-*-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cuda-toolkit-12-4

# 显式将 CUDA 12.4 的 lib64 库路径优先加载，防止链接器调用系统的旧版库，以避免 undefined reference to cublasSetWorkspace_v2
export LD_LIBRARY_PATH=/usr/local/cuda-12.4/lib64:$LD_LIBRARY_PATH
export PATH=/usr/local/cuda-12.4/bin:$PATH
export CUDA_PATH=/usr/local/cuda-12.4

rm -rf build
CUDACXX=/usr/local/cuda-12.4/bin/nvcc cmake -B build -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=86 -DLLAMA_BUILD_TESTS=OFF -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-12.4
cmake --build build --config Release -j$(nproc)