Installing SpanMarker

You may install the span_marker Python module via pip like so:

pip install span_marker

PyTorch GPU support

If you are installing SpanMarker locally, you may wish to install torch in such a way that PyTorch models can be executed on the GPU. This generally results in large speed improvements, both for training and predicting entities. I recommend following the official PyTorch installation guide.

Once installed, you can verify whether torch is compiled with CUDA support like so:

>>> import torch
>>> torch.cuda.is_available()
True

And in the context of SpanMarker, you can always move a model to CUDA using:

>>> model = SpanMarkerModel.from_pretrained(...)
>>> model.cuda()
>>> model.device
device(type='cuda', index=0)