textgenrnn

Why we included this project

textgenrnn wraps a Keras and TensorFlow recurrent network behind a small Python API, so you can train a char- or word-level text generator without wiring up attention layers or bidirectional cells yourself. Models trained on a GPU will run on a plain CPU later, and the included pretrained checkpoint is only a couple of megabytes, so you can fine-tune on a new corpus in a single call. The result is a quick way to turn a pile of posts, comments, or logs into synthetic text that sounds plausibly like the source. The interactive mode, which suggests likely next characters, makes it easy to steer generations by hand instead of just sampling blindly.