lucasjinreal/tensorflow_poems

LiBai AI Composer is a Python project that trains a neural network to write classical Chinese poetry. Named after the Tang dynasty poet Li Bai, the model learns from a corpus of ancient Chinese poems and then generates new ones in a similar style. The README shows example output, a seven-line verse the model produced after training on roughly 40,000 Tang poems. The project uses TensorFlow and an LSTM (a type of recurrent neural network well suited for generating sequences of text). Training runs by executing a single Python script, and the model comes with preprocessed data included, so new users do not need to collect or format the source poems themselves. A separate script handles poem generation after training is complete. Aside from classical poetry, the project was extended to also imitate the lyrics of a popular Chinese pop musician. The README notes that this feature needs more training data to produce convincing results and invites contributors to submit song lyric text files. The repository also includes acrostic poem support, where the first character of each line spells out a word of your choosing. Users pass a starting character or word and the model constructs a poem around it. The README is written primarily in Chinese. The project is open source under the Apache license. It was built as a learning exercise and has received several updates since 2017 to fix bugs like an infinite loop during generation and to simplify the data preprocessing steps.