This repository contains State of the Art Language models and Classifier for Urdu, spoken mainly in Pakistan and India, and also in Nepal, Bangladesh and several other countries.
The models trained here have been used in Natural Language Toolkit for Indic Languages (iNLTK)
| Architecture/Dataset | Urdu Wikipedia Articles |
|---|---|
| ULMFiT | 13.19 |
| TransformerXL | 12.55 |
| Dataset | Accuracy | Kappa Score |
|---|---|---|
| Urdu News Dataset | 95.28 | 91.58 |
| Architecture | Visualization |
|---|---|
| ULMFiT | Embeddings projection |
| TransformerXL | Embeddings projection |
Download pretrained ULMFiT LM from here
Download pretrained TransformerXL LM from here
Download classifier from here
Trained tokenizer using Google's sentencepiece
Download the trained model and vocabulary from here