metadata
language:
- en
pipeline_tag: text2text-generation
metrics:
- f1
tags:
- grammatical error correction
- GEC
- russian
This is a fine-tuned version of Multilingual Bart trained on Russian in particular on the public dataset RULEC-GEC for Grammatical Error Correction.
To initialize the model:
from transformers import MBartForConditionalGeneration, MBart50TokenizerFast
model = MBartForConditionalGeneration.from_pretrained("MRNH/mbart-russian-grammar-corrector")
To generate text using the model:
tokenizer = MBart50TokenizerFast.from_pretrained("MRNH/mbart-russian-grammar-corrector", src_lang="ru_RU", tgt_lang="ru_RU")
input = tokenizer("I was here yesterday to studying",text_target="I was here yesterday to study", return_tensors='pt')
output = model.generate(input["input_ids"],attention_mask=input["attention_mask"],forced_bos_token_id=tokenizer_it.lang_code_to_id["ru_RU"])