This model is trained on Roman Urdu POS dataset of 100000 tokens.