dreiss/java-demo

PyTorch Java demo

java-demo's Issues

Does libtorch-java support CUDA?

The demo you provide is very efficient, and I have traced my model to inference with JAVA, my question is whether I could use CUDA to inference the model, since the java method of libtorch is far less than C++, and I didn't find the to(kCUDA) on https://pytorch.org/javadoc/.

Looking forward to your reply.

Extracting extra-file information. Where is the ExtraFilesMap()?

This is a fantastic library - I've been testing it out and great to have native java implementation which works with existing software. However I've come across an issue I can't seem to resolve. I am unable to extract any of the metadata associated with a pre-trained model?

When saving a trained model, extra information can be added in Python or C++ using an ExtraFilesMap object. In this case the extra info is the type of audio data the model was trained on; this is vital for pre-conditioning audio data before it's passed to a model for classification. In Python this extra metadata can be extracted from a saved model via...

extra_files = torch._C.ExtraFilesMap()
model = torch.jit.load(filepath_model, _extra_files=extra_files)
dataOptsStr = extra_files['dataOpts']

In Java a trained model can be loaded using...

Module mod = Module.load(modelPath);

However, there does not appear to be any method to extract the ExtraFileMap information. I've spent a long time trying to find a solution with no luck. Am missing something here? It seems like important functionality for sharing models between different languages etc.?

Thanks for the help.

no prob

When windows version born ?

Hope as soon as possible...

java-cpu-only

Firstly, thank you for building such a great tool.
But how can I build a version which using cpu only because file libtorch_cuda.so is too big.

Thank you.