Documenting the GPU inference feature.

README.md

@@ -56,7 +56,9 @@ Alternatively, you can select a specific model by defining LAMBO variant (`LAMBO
 lambo = Lambo.get('LAMBO-UD_Polish-PDB')
 ```
 
-Finally, if a subword splitter is available, you can opt out of using it by providing `with_splitter=False` as an argument to the `get()` function.
+There are two optional arguments to the `get()` function:
+- You can opt out of using subword splitter by providing `with_splitter=False`.
+- You can point to a specific pyTorch device by providing `device` parameter, for example `device=torch.device('cuda')` to enable GPU acceleration.
 
 Once the model is ready, you can perform segmentation of a given text:
 ```