app.py

import wikipedia as wiki
import pprint as pp
from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline, AutoModelForSeq2SeqLM
import torch
import gradio as gr

def greet(name):
    #question = 'Why is the sky blue?'
    question = name

    
    results = wiki.search(question)
    #print("Wikipedia search results for our question:\n")
    #pp.pprint(results)
    
    page = wiki.page(results[0])
    text = page.content
    #print(f"\nThe {results[0]} Wikipedia article contains {len(text)} characters.")
    
    #print(text)
    
    
    model_name = "deepset/roberta-base-squad2"
    
    #from transformers import AutoModel
    
    #model_name = AutoModelForQuestionAnswering.from_pretrained('./roberta-base-squad2/')
    
    def get_sentence(text, pos):
        start = text.rfind('.', 0, pos) + 1
        end = text.find('.', pos)
        if end == -1:
            end = len(text)
        return text[start:end].strip()
    
    
    # a) Get predictions
    nlp = pipeline('question-answering', model=model_name, tokenizer=model_name)
    QA_input = {
        'question': question,
        'context': text
    }
    res = nlp(QA_input)
    
    # b) Load model & tokenizer
    model = AutoModelForQuestionAnswering.from_pretrained(model_name)
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    #print("{'answer': '"+res['answer']+"', 'text': '")
    #print(res['answer'])
    #print("', 'text': '")
    
    position = res['start']
    #words = sum(map(str.split, text), [])
    #sentence = ' '.join(words[position-1:]).split('.')[0] + '.'
    
    #print(get_sentence(text, position)+'.')
    
    tokenizer = AutoTokenizer.from_pretrained("tuner007/pegasus_paraphrase")
    
    model = AutoModelForSeq2SeqLM.from_pretrained("tuner007/pegasus_paraphrase")
    
    torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'
    
    def get_response(input_text,num_return_sequences,num_beams):
      batch = tokenizer([input_text],truncation=True,padding='longest',max_length=60, return_tensors="pt").to(torch_device)
      translated = model.generate(**batch,max_length=60,num_beams=num_beams, num_return_sequences=num_return_sequences, temperature=1.5)
      tgt_text = tokenizer.batch_decode(translated, skip_special_tokens=True)
      return tgt_text
    
    
    num_beams = 20
    num_return_sequences = 1
    context = get_sentence(text, position)+'.'
    #print(get_response(context,num_return_sequences,num_beams)[0])
    #print("'}")
    return "{'answer': '"+res['answer']+"', 'text': '"+get_response(context,num_return_sequences,num_beams)[0]+"'}"

demo = gr.Interface(fn=greet, inputs="text", outputs="text")

demo.launch()