Dmitry Alexeev

@vectorflux should be fixed. You can't allocate base class in nvhpc compiler and use it as derived in the later present statements. I'm checking with Jeff Larkin if that's standard-conforming,...

Hi @ch21d012 , sorry for the delay. Please have a look at the resnet example, I hope it should do what you want: https://github.com/alexeedm/pytorch-fortran/blob/main/examples/resnet_forward/resnet_forward.f90 Let me know if there are...

Right. It seems that there is an error loading the model. Are you sure that the file `../examples/my_model/model_0.005_150_0.97_0.pth` exists and readable? Does the original resnet example work?

Ok, can you show me your fortran code? And would it be possible to share the model file (possibly privately?) By the way, how did you save your model into...

Ok, I see a pontential problem, maybe you don't even have to share the files. Before calling `torch.save` you need to trace or JIT your model into a TorchScript: ```...

@ch21d012 I've just pushed a few changes that should improve error reporting, at least for GCC. If you're still seeing any issues, please try to pull the latest `main` branch,...

Hi @pimbcodes, thanks! The error seems to happen when the code tries to run the Python interpreter. Are you sure you're passing the correct module file name when running the...

@ch21d012 could you give a bit more details on how you've compiled the code (compiler, container or not, which compile script you've used, etc.)? The code fails because it needs...

I'm having quite a lot of other work recently, I'm sorry for taking longer folks! I'll come back here as soon as I can

@ch21d012 you can definitely build the code without CUDA, look at `make_gnu.sh` script for example. The main thing is to install cpu-only pytorch package (check on their web-site) and set...