按照https://www.mindspore.cn/lite/docs/zhCN/r2.0/use/build.html#%E6%A8%A1%E5%9D%97%E6%9E%84%E5%BB%BA%E7%BC%96%E8%AF%91%E9%80%89%E9%A1%B9
在Linux上编译mindspore_lite的时候，已经在编译前export MSLITE_ENABLE_CONVERT_PYTORCH_MODEL=on。转换前加入libtorch的环境变量：export LD_LIBRARY_PATH="/home/tt/libtorch/lib:${LD_LIBRARY_PATH}" && export LIB_TORCH_PATH="/home/tt/libtorch"。

执行的命令是 bash build.sh -I x86_64 -j2

编译成功后
使用./converter_lite --fmk=PYTORCH --modelFile=hwnum_r0329_2.pt --outputFile=model
显示Unsupported to converter models with fmk: 5

mindspore版本编译出来的工具是mindspore-lite-2.0.1-linux-x64.tar.gz, 类型是CPU， 环境ubuntu18.04LTS, gcc7.5.0, cmake3.18.3, python3.8, OpenHarmony_NDK 3.2release, AndroidNDK=r20

执行grep -r "MSLITE_ENABLE_CONVERT_PYTORCH_MODEL" . 结果是

./mindspore/lite/CMakeLists.txt:option(MSLITE_ENABLE_CONVERT_PYTORCH_MODEL "enable to convert pytorch model" on)
./mindspore/lite/CMakeLists.txt:if(DEFINED ENV{MSLITE_ENABLE_CONVERT_PYTORCH_MODEL} AND DEFINED ENV{LIB_TORCH_PATH})
./mindspore/lite/CMakeLists.txt:set(ENABLE_CONVERT_PYTORCH_MODEL $ENV{MSLITE_ENABLE_CONVERT_PYTORCH_MODEL})
./mindspore/lite/cmake/lite_options.cmake:option(MSLITE_ENABLE_CONVERT_PYTORCH_MODEL "enable to convert pytorch model" on)
./mindspore/lite/cmake/lite_options.cmake:if(DEFINED ENV{MSLITE_ENABLE_CONVERT_PYTORCH_MODEL} AND DEFINED ENV{LIB_TORCH_PATH})
./mindspore/lite/cmake/lite_options.cmake: set(ENABLE_CONVERT_PYTORCH_MODEL $ENV{MSLITE_ENABLE_CONVERT_PYTORCH_MODEL})
./mindspore/lite/build/CMakeCache.txt:MSLITE_ENABLE_CONVERT_PYTORCH_MODEL:BOOL=ON

./mindspore/lite/CMakeLists.txt 中我修改了off为on，第一次编译之后没有效果， 第二次编译前加上了export MSLITE_ENABLE_CONVERT_PYTORCH_MODEL=on， 仍然显示不支持的格式。