arijitx/IndicBART-bn-QuestionGeneration

Intro

Trained on IndicNLGSuit IndicQuestionGeneration data for Bengali the model is finetuned from IndicBART

Finetuned Command

python run_summarization.py --model_name_or_path bnQG_models/checkpoint-32000 --do_eval --train_file train_bn.json 
--validation_file valid_bn.json --output_dir bnQG_models --overwrite_output_dir --per_device_train_batch_size=2 
--per_device_eval_batch_size=4 --predict_with_generate --text_column src --summary_column tgt --save_steps 4000 
--evaluation_strategy steps --gradient_accumulation_steps 4 --eval_steps 1000 --learning_rate 0.001 --num_beams 4 
--forced_bos_token "<2bn>" --num_train_epochs 10 --warmup_steps 10000

Sample Line from train data

{"src": "प्राणबादी [SEP] अर्थाॎ, तिनि छिलेन एकजन सर्बप्राणबादी। </s> <2bn>", "tgt": "<2bn> कोन दार्शनिक दृष्टिभङ्गि ओय़ाइटजेर छिल? </s>"}

Inference

script = "সুভাষ ১৮৯৭ খ্রিষ্টাব্দের ২৩ জানুয়ারি ব্রিটিশ ভারতের অন্তর্গত বাংলা প্রদেশের উড়িষ্যা বিভাগের (অধুনা, ভারতের ওড়িশা রাজ্য) কটকে জন্মগ্রহণ করেন।"
answer = "১৮৯৭ খ্রিষ্টাব্দের ২৩ জানুয়ারি"
inp = answer +" [SEP] "+script + " </s> <2bn>"
inp_tok = tokenizer(inp, add_special_tokens=False, return_tensors="pt", padding=True).input_ids
model.eval() # Set dropouts to zero

model_output=model.generate(inp_tok, use_cache=True, 
                            num_beams=4, 
                            max_length=20, 
                            min_length=1, 
                            early_stopping=True, 
                            pad_token_id=pad_id, 
                            bos_token_id=bos_id, 
                            eos_token_id=eos_id, 
                            decoder_start_token_id=tokenizer._convert_token_to_id_with_added_voc("<2bn>")
                        )
decoded_output=tokenizer.decode(model_output[0], skip_special_tokens=True, clean_up_tokenization_spaces=False)

Citations

@inproceedings{dabre2021indicbart,
    title={IndicBART: A Pre-trained Model for Natural Language Generation of Indic Languages}, 
    author={Raj Dabre and Himani Shrotriya and Anoop Kunchukuttan and Ratish Puduppully and Mitesh M. Khapra and Pratyush Kumar},
    year={2022},
    booktitle={Findings of the Association for Computational Linguistics},
    }    


@misc{kumar2022indicnlg,
  title={IndicNLG Suite: Multilingual Datasets for Diverse NLG Tasks in Indic Languages}, 
  author={Aman Kumar and Himani Shrotriya and Prachi Sahu and Raj Dabre and Ratish Puduppully and Anoop Kunchukuttan and Amogh Mishra and Mitesh M. Khapra and Pratyush Kumar},
  year={2022},
  eprint={2203.05437},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}