GPU version Compilation Error

[zxyf1]

I was trying to use dfLowMachFoam GPU version with latest master commit. The code before modifying was fine to compile and run, but with only CPU functionality. Then I tried to switch on GPU functionality by uncomment the macro "#define GPUSolver_" in the dfLowMachFoam.C. Then I ran the command ".. install.sh". However` I occurred compilation error:

/usr/bin/ld.bfd: cannot find /opt/deepflame-dev/src/dfChemistryModel/DNNInferencer/build/libDNNInferencer.so: No such file or directory
collect2: error: ld returned 1 exit status

Then I tried to compile DNNInference independently in deepflame-dev/src/dfChemistryModel/DNNInferencer/ independently, but still there is error indicating DNNInference compilation is not completed which requires libtorch.

Therefore, I was wondering whether it is neccessary to compile GPU version with libtorch，or I can use no torch version?
Please tell me if there is any available GPU version code I can use without modification.

[zxyf1]

I also tried to use libtorch for compilation, but occurred version compatibility error. Therefore, please tell me which version of libtorch did you use in GPU version.