metadata

license: apache-2.0
language:
  - gu

Model description

The model is a mt5-small version of Gujarati Grammarly for spell correction given a sentence. Only this small version checkpoints are open source.

Example usage:

from transformers import AutoTokenizer
import tensorflow as tf
from transformers import TFAutoModelForSeq2SeqLM
from transformers import create_optimizer

model_checkpoint = "Jayveersinh-Raj/guj-grammar-small"
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
model = TFAutoModelForSeq2SeqLM.from_pretrained(model_checkpoint)
 
sent="સુંદરકાંડના પ્રારંભમાં હનૂમાન બળવાન તો છે પણ સાથે-સાથે બુદ્ધિમાન પણ છે તેની રોચક ધર્મકથા છૈ"
inputs = tokenizer.encode(sent, return_tensors='tf')
output_ids = model.generate(inputs, max_length=128, num_beams = 4, early_stopping=True)
output = tokenizer.decode(output_ids[0], skip_special_tokens=True)

print("Generated Correction:")
print(output)

Notes:

Only supports Gujarati language for now
Private dataset is used
Only Tensorflow model is available for now, Pytorch checkpoints would be available soon.