gector

Why we included this project

If you're building grammar checking into an editor or writing tool, GECToR is worth studying. It's the official PyTorch code for a tagging-based approach to grammatical error correction: instead of rewriting the whole sentence, the model assigns each token a small edit, which keeps inference cheap enough to run at scale. The repo includes a full training pipeline and pretrained checkpoints for BERT, RoBERTa, and XLNet that score well on standard GEC benchmarks, so you can load those weights for immediate inference or fine-tune them on your own text. It also ships the related text-simplification-by-tagging model behind the same interface, covering two editing tasks from one codebase. The preprocessing scripts and the two-stage fine-tuning recipe give you a concrete baseline to compare against your data before you bet on a production writing-assistant stack.