I am using the following model facebook/mbart-large-50-many-to-many-mmt from hugginface to perform machine translation. However, I cannot perform multilingual translation (i.e., Hindi to Marathi, french to german, for example). If I try to do so (take Hindi to Marathi, for example), the output upon decoding is in the English language. Please tell me how to convert from language A to B when neither of them is the English language.
Thanks
Code snippet used
from transformers import AutoTokenizer, MBartForConditionalGeneration
import torch
model = MBartForConditionalGeneration.from_pretrained("facebook/mbart-large-50-many-to-many-mmt")
tokenizer = AutoTokenizer.from_pretrained("facebook/mbart-large-50-many-to-many-mmt")
article_hi = "संयुक्त राष्ट्र के प्रमुख का कहना है कि सीरिया में कोई सैन्य समाधान नहीं है"
# article_en = "Hello World!"
tokenizer.src_lang = "hi_IN"
encoded_hi = tokenizer(article_hi, return_tensors="pt")
encoded_hi = encoded_hi.to(torch.device("cuda"))
model = model.to(torch.device("cuda"))
generated_tokens = model.generate(**encoded_hi, forced_bos_token_id=tokenizer.lang_code_to_id["mr_IN"])
print(tokenizer.decode(generated_tokens[0], skip_special_tokens=True))