This is the IndicBART model. For detailed documentation look here: https://indicnlp.ai4bharat.org/indic-bart/ and https://github.com/AI4Bharat/indic-bart/
Usage:
from transformers import MBartForConditionalGeneration, AutoModelForSeq2SeqLM
from transformers import AlbertTokenizer, AutoTokenizer
tokenizer = AlbertTokenizer.from_pretrained("prajdabre/IndicBARTTokenizer", do_lower_case=False, use_fast=False, keep_accents=True)
# Or use tokenizer = AutoTokenizer.from_pretrained("prajdabre/IndicBARTTokenizer", do_lower_case=False, use_fast=False, keep_accents=True)
model = MBartForConditionalGeneration.from_pretrained("prajdabre/IndicBART")
# Or use model = AutoModelForSeq2SeqLM.from_pretrained("prajdabre/IndicBART")
# First tokenize the input and outputs. The format below is how IndicBART was trained so the input should be "Sentence </s> <2xx>" where xx is the language code. Similarly, the output should be "<2yy> Sentence </s>".
inp = tokenizer("I am a boy <\/s> <2en>", add_special_tokens=False, return_tensors="pt", padding=True).input_ids
out = tokenizer("<2hi> मैं एक लड़का हूँ <\/s>", add_special_tokens=False, return_tensors="pt", padding=True).input_ids
model_outputs=model(input_ids=inp, decoder_input_ids=out[:,0:-1], labels=out[:,1:])
# For loss
model_outputs.loss ## This is not label smoothed.
# For logits
model_outputs.logits
# For generation. Pardon the messiness. Note the decoder_start_token_id.
model.eval() # Det dropouts to zero
model_output=model.generate(inp, use_cache=True, num_beams=4, max_length=20, min_length=1, early_stopping=True, pad_token_id=tokenizer.pad_token_id, decoder_start_token_id=tokenizer(["<2en>"], add_special_tokens=False).input_ids[0][0])
# Decode to get output strings
decoded_output=tokenizer.decode(model_output[0], skip_special_tokens=True, clean_up_tokenization_spaces=False)
print(decoded_output) # I am a boy
# What if we mask?
inp = tokenizer("I am [MASK] </s> <2en>", add_special_tokens=False, return_tensors="pt", padding=True).input_ids
model_output=model.generate(inp, use_cache=True, num_beams=4, max_length=20, min_length=1, early_stopping=True, pad_token_id=tokenizer.pad_token_id, decoder_start_token_id=tokenizer(["<2en>"], add_special_tokens=False).input_ids[0][0])
decoded_output=tokenizer.decode(model_output[0], skip_special_tokens=True, clean_up_tokenization_spaces=False)
print(decoded_output) # I am happy
Notes:
- This is compatible with the latest version of transformers but was developed with version 4.3.2 so consider using 4.3.2 if possible.
- The tokenizer repo is kept separate from the model repo because unlike mBART-25 and mBART-50 which use a BPE model (MBartTokenizer class) whereas we use the sentencepiece model (AlbertTokenizer class).
- Currently, keeping the tokenizer and model files in the same repo complicates things so keeping them separate is a temporary solution. This will be fixed in future versions.
- While I have only shown how to let logits and loss and how to generate outputs, you can do pretty much everything the MBartForConditionalGeneration class can do as in https://huggingface.co/docs/transformers/model_doc/mbart#transformers.MBartForConditionalGeneration