app.py

import gradio as gr
from transformers import pipeline
from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras.preprocessing.text import text_to_word_sequence
import pickle
import re
from tensorflow.keras.models import load_model

# Load long model
with open('lstm-qa-long-answers-model/tokenizer.pickle', 'rb') as handle:
    tokenizer = pickle.load(handle)
long_answer_model = load_model('lstm-qa-long-answers-model/model.h5')

def clean_text(text):
    text = re.sub(r'<.*?>', '', text)  
    text = re.sub(r'\[\d+\]', '', text)  
    text = re.sub(r'[^a-zA-Z0-9\s().,]', '', text) 
    return text

def remove_parentheses(text):
    pattern = r'\([^)]*\)'
    return re.sub(pattern, '', text)

def predict_correct_answer(question, answer1, answer2):
    answers = [answer1, answer2]
    correct_answer = None
    best_score = 0
    for answer in answers:
        clean_answer = clean_text(answer)
        question_seq = tokenizer.texts_to_sequences([question])
        answer_seq = tokenizer.texts_to_sequences([clean_answer])
        padded_question = pad_sequences(question_seq, padding='post')
        padded_answer = pad_sequences(answer_seq, maxlen=300, padding='post', truncating='post')
        score = long_answer_model.predict([padded_answer, padded_question])[0][0]
        if score > best_score:
            best_score = score
            correct_answer = clean_answer

    return correct_answer, best_score

# def split_into_sentences(text):
#     sentences = re.split(r'\.\s*', text)
#     return [s.strip() for s in sentences if s]

# def predict_answer(context, question):
#     sentences = split_into_sentences(context)
#     best_sentence = None
#     best_score = 0

#     for sentence in sentences:
#         clean_sentence = clean_text(sentence)
#         question_seq = tokenizer.texts_to_sequences([question])
#         sentence_seq = tokenizer.texts_to_sequences([clean_sentence])

#         max_sentence_length = 300
#         padded_question = pad_sequences(question_seq, padding='post')
#         padded_sentence = pad_sequences(sentence_seq, maxlen=max_sentence_length, padding='post', truncating='post')

#         score = long_answer_model.predict([padded_sentence, padded_question])[0]

#         if score > best_score:
#             best_score = score
#             best_sentence = clean_sentence

#     return best_score, best_sentence

# Load short model
distilbert_base_uncased = pipeline(model="Nighter/QA_wiki_data_short_answer", from_tf=True)
bert_base_uncased  = pipeline(model="Nighter/QA_bert_base_uncased_wiki_data_short_answer", from_tf=True)
roberta_base = pipeline(model="Nighter/QA_wiki_data_roberta_base_short_answer", from_tf=True)
longformer_base = pipeline(model="aware-ai/longformer-squadv2")

# Function to answer on all models
def answer_questions(context, question):
    # long_score, long_answer = predict_answer(context, question)
    distilbert_base_uncased_result = distilbert_base_uncased(question=question, context=remove_parentheses(context))
    bert_base_uncased_result = bert_base_uncased(question=question, context=remove_parentheses(context))
    roberta_base_result = roberta_base(question=question, context=remove_parentheses(context))
    longformer_base_result = longformer_base(question=question, context=remove_parentheses(context))
    return distilbert_base_uncased_result['answer'], distilbert_base_uncased_result['score'], bert_base_uncased_result['answer'], bert_base_uncased_result['score'], roberta_base_result['answer'], longformer_base_result['score'], longformer_base_result['answer'], roberta_base_result['score'] #, long_answer, long_score

# App Interface
with gr.Blocks() as app:
    gr.Markdown("<center> <h1>Question Answering with Short and Long Answer Models </h1> </center><hr>")
    with gr.Tab("QA Short Answer"):
        with gr.Row():
            with gr.Column():
                context_input = gr.Textbox(lines=8, label="Context", placeholder="Input Context here...") 
                question_input = gr.Textbox(lines=3, label="Question", placeholder="Input Question here...")
                submit_btn = gr.Button("Submit")
                gr.ClearButton([context_input,question_input])
            with gr.Column():
                with gr.Row():
                    with gr.Column(scale=6):
                        distilbert_base_uncased_output = gr.Textbox(lines=2, label="Distil BERT Base Uncased")
                    with gr.Column(scale=2):
                        distilbert_base_uncased_score = gr.Number(label="Distil BERT Base Uncased Score")
                with gr.Row():
                    with gr.Column(scale=6):
                        bert_base_uncased_output = gr.Textbox(lines=2, label="BERT Base Uncased")
                    with gr.Column(scale=2):
                        bert_base_uncased_score = gr.Number(label="BERT Base Uncased Score")
                with gr.Row():
                    with gr.Column(scale=6):
                        roberta_base_output = gr.Textbox(lines=2, label="RoBERTa Base")
                    with gr.Column(scale=2):
                        roberta_base_score = gr.Number(label="RoBERTa Base Score")
                with gr.Row():
                    with gr.Column(scale=6):
                        longformer_base_output = gr.Textbox(lines=2, label="Longformer Base")
                    with gr.Column(scale=2):
                        longformer_base_score = gr.Number(label="Longformer Base Score")

        submit_btn.click(fn=answer_questions, inputs=[context_input, question_input], outputs=[distilbert_base_uncased_output, distilbert_base_uncased_score, bert_base_uncased_output, bert_base_uncased_score, roberta_base_output, roberta_base_score, longformer_base_output, longformer_base_score])
        examples='examples'
        gr.Examples(examples,[context_input, question_input],[distilbert_base_uncased_output, distilbert_base_uncased_score, bert_base_uncased_output, bert_base_uncased_score, roberta_base_output, roberta_base_score, longformer_base_output, longformer_base_score],answer_questions)
        
    with gr.Tab("Long Answer Prediction"):
        with gr.Row():
            with gr.Column():
                long_question_input = gr.Textbox(lines=3,label="Question", placeholder="Enter the question")
                answer1_input = gr.Textbox(lines=3,label="Answer 1", placeholder="Enter answer 1")
                answer2_input = gr.Textbox(lines=3,label="Answer 2", placeholder="Enter answer 2")
                submit_btn_long = gr.Button("Submit")
                gr.ClearButton([long_question_input, answer1_input, answer2_input])

            with gr.Column():
                correct_answer_output = gr.Textbox(lines=3,label="Correct Answer")
                score_output = gr.Number(label="Score")
    
        submit_btn_long.click(fn=predict_correct_answer, inputs=[long_question_input, answer1_input, answer2_input], 
                         outputs=[correct_answer_output, score_output])
        
        long_examples = 'long_examples'
        gr.Examples(long_examples,[long_question_input, answer1_input, answer2_input])
                    
if __name__ == "__main__":
    app.launch()