TextGAN-PyTorch

Why we included this project

Researchers and graduate students working on adversarial text generation will find this a compact PyTorch codebase with the classic GAN methods for discrete sequence data, including SeqGAN, LeakGAN, MaliGAN, RelGAN, and DPGAN, each tied to its original paper and accompanied by a file describing the training loop. It also includes category-conditioned variants like SentiGAN and CatGAN, plus evaluation metrics and logging, so you can run baselines and compare results on a single harness. Since most of these models were released only as TensorFlow code, this repo saves PyTorch users the trouble of porting them. One honest caveat: the methods predate the large-language-model era, so the real value now is as a benchmarking and educational tool rather than a production text generator. If you need a reference implementation or a ready baseline set for a thesis chapter, this is a sensible place to start.