Grammar Error Correction
A grammar error correction model for Polish, published on Hugging Face. It finds grammatical errors in Polish text and proposes corrections — built as a token-classification pipeline in the GECToR style, on top of the HerBERT-large encoder (allegro/herbert-large-cased).
What it does
Given a Polish sentence, the model tags the tokens that need correcting and produces the corrected forms. The pipeline follows the GECToR pattern — grammar error correction treated as a sequence-tagging task — and the weights are exported in both ONNX and safetensors formats.
Where to find it
- Grammar error correction model on Hugging Face — the model card and weights
- citan’s Hugging Face profile — hub page that links both models
Technical Details
- Task: token-classification (GECToR-style grammar error correction)
- Base model: HerBERT (allegro/herbert-large-cased)
- Exports: ONNX and safetensors
- License: CC-BY-NC-4.0