uzw/bart-large-question-generation

Generating Questions Given Context and Answers

Traditional BART model is not pre-trained on QG tasks. We fine-tuned facebook/bart-large model using 55k human-created question answering pairs with contexts collected by Demszky et al. (2018). The dataset includes SQuAD and QA2D question answering pairs associated with contexts.

How to use

Here is how to use this model in PyTorch:

from transformers import BartForConditionalGeneration, BartTokenizer
import torch

tokenizer = BartTokenizer.from_pretrained('uzw/bart-large-question-generation')
model = BartForConditionalGeneration.from_pretrained('uzw/bart-large-question-generation')

context = "The Thug cult resides at the Pankot Palace."
answer = "The Thug cult"

inputs = tokenizer.encode_plus(
    context, 
    answer, 
    max_length=512, 
    padding='max_length', 
    truncation=True, 
    return_tensors='pt'
)

with torch.no_grad():
    generated_ids = model.generate(
        input_ids=inputs['input_ids'], 
        attention_mask=inputs['attention_mask'],
        max_length=64,  # Maximum length of generated question
        num_return_sequences=3,  # Generate multiple questions
        do_sample=True,  # Enable sampling for diversity
        temperature=0.7  # Control randomness of generation
    )

generated_questions = tokenizer.batch_decode(
    generated_ids, 
    skip_special_tokens=True
)

for i, question in enumerate(generated_questions, 1):
    print(f"Generated Question {i}: {question}")

Adjusting parameter num_return_sequences to generate multiple questions.