this-word-does-not-exist

Why we included this project

This project fine-tunes a GPT-2 variant to invent words from scratch, complete with plausible definitions and example sentences. The repo ships both halves of the pipeline: pretrained checkpoints and a WordGenerator class for CPU inference, plus the training scripts and dictionary extraction code used to build those models. That split makes it a solid learning artifact. You can generate words immediately or follow the notebooks and train scripts to reproduce the whole process on your own dictionary source. It is a good fit for developers curious about domain-specific fine-tuning, small teams poking at constrained text generation, and educators who want an approachable end-to-end language model example.