seq2seq-couplet

Why we included this project

Chinese couplets are a good test of a sequence model: the second line has to echo the first in meaning while obeying strict tonal and character-count rules. This project trains a TensorFlow seq2seq model on a couplet dataset to do exactly that, producing a matching 下联 from a given 上联. The repo bundles the training code, a web server for serving the trained model, and a Dockerfile, so you can go from dataset to a playable demo without much setup. The sample outputs show the model turning out coherent, poetic responses rather than random character sequences. It's a handy reference if you want to study sequence-to-sequence architectures on a small, well-defined creative task, or just want a fun way to see what deep learning does with traditional Chinese poetry。