Work In Progress.
An unofficial implementation of R-net in PyTorch.
Natural Language Computing Group, MSRA: R-NET: Machine Reading Comprehension with Self-matching Networks
See: https://www.microsoft.com/en-us/research/publication/mrc/
Python 3.5/3.6 and PyTorch 0.2
Current Problems
It seems it is impossible to train the model on single GPU because it is a lack of memory for a reasonable batch size. The next stage is to modify the code to train the model with mutiple GPU (e.g. 1 GPU per layer)