CopyNet enhances the capability of standard encoder-decoder models, allowing them to generate output sequences that include "out of vocabulary" tokens present in the input sequence.

This project is a refined re-implementation of CopyNet, which originally inspired from code by adamklec. This version introduces several improvements over the original:

• A completely re-engineered dataset class that is abstracted from file formats, thus making it simpler to load your custom data.
• Integration with pretrained word embeddings (demonstrated using GLOVE), which facilitates improved model performance.

Each line of input is composed of two sequences: the first represents the input sequence, while the second acts as the target output sequence. Tokens in each sequence should be separated by spaces.

To utilize this implementation, follow the instructions provided in playground.ipynb in sequence.