Ubuntu20.04安装ktransformers 卡在subprocess.CalledProcessError: Command xxxx returned non-zero exit sta

系统环境 ubuntu20.04 

安装ktransformers ，运行install.sh的时候卡在编译的一步

subprocess.CalledProcessError: Command '['cmake', '/home/fox/env/kt31/ktransformers/csrc/ktransformers_ext', '-DCMAKE_LIBRARY_OUTPUT_DIRECTORY=/home/fox/env/kt31/ktransformers/build/lib.linux-x86_64-cpython-311/', '-DPYTHON_EXECUTABLE=/home/fox/.conda/envs/kt3/bin/python3.11', '-DCMAKE_BUILD_TYPE=Release', '-DKTRANSFORMERS_USE_CUDA=ON', '-D_GLIBCXX_USE_CXX11_ABI=0', '-DLLAMA_NATIVE=ON', '-DEXAMPLE_VERSION_INFO=0.3.1+cu118torch23avx2']' returned non-zero exit status 1.

让AI把这段编译的代码提出来

cmake \
  /home/fox/env/kt31/ktransformers/csrc/ktransformers_ext \
  -DCMAKE_LIBRARY_OUTPUT_DIRECTORY=/home/fox/env/kt31/ktransformers/build/lib.linux-x86_64-cpython-311/ \
  -DPYTHON_EXECUTABLE=/home/fox/.conda/envs/kt3/bin/python3.11 \
  -DCMAKE_BUILD_TYPE=Release \
  -DKTRANSFORMERS_USE_CUDA=ON \
  -D_GLIBCXX_USE_CXX11_ABI=0 \
  -DLLAMA_NATIVE=ON \
  -DEXAMPLE_VERSION_INFO=0.3.1+cu118torch23avx2

手动运行debug 发现报错：

CMake Error at CMakeLists.txt:316 (find_package):
  By not providing "FindCUDAToolkit.cmake" in CMAKE_MODULE_PATH this project
  has asked CMake to find a package configuration file provided by
  "CUDAToolkit", but CMake did not find one.

  Could not find a package configuration file provided by "CUDAToolkit" with
  any of the following names:

    CUDAToolkitConfig.cmake
    cudatoolkit-config.cmake

  Add the installation prefix of "CUDAToolkit" to CMAKE_PREFIX_PATH or set
  "CUDAToolkit_DIR" to a directory containing one of the above files.  If
  "CUDAToolkit" provides a separate development package or SDK, be sure it
  has been installed.

解决方案：

查阅资料，排除gcc g++版本问题后(cuda12,不支持老版本gcc 需要使用gcc版本13，cuda11.8，可以使用gcc版本11)，遂检查cmake版本，ubuntu20.04自带的cmake版本3.16，升级成 3.20，问题解决

cmake升级教程参考：Ubuntu20.04升级cmake（不用去额外删除现有版本）_cmake 3.20 or higher is required. you are running -CSDN博客

其他的坑点：

如果想使用cuda11.8，需要手动安装 pytorch2.3.0 然后再去ktransformers 文件夹中 bash instal.sh

# CUDA 11.8

conda install pytorch==2.3.0 torchvision==0.18.0 torchaudio==2.3.0 pytorch-cuda=11.8 -c pytorch -c nvidia