CMAKE_ARGS="-DLLAMA_CUDA=on" pip install llama-cpp-python
pip install sentencepiece
pip install lmql[hf]

