app.py

import streamlit as st
from datasets import load_dataset

# from transformers import T5Tokenizer, T5ForConditionalGeneration
from transformers import AutoTokenizer, AutoModelForQuestionAnswering, pipeline, AutoModelForQuestionAnswering
import torch
model_path = "./kaggle-3/working/bert_qa"
# model_path = "./flan_t5_qa_20"

tokenizer_new = AutoTokenizer.from_pretrained(model_path)
model_new = AutoModelForQuestionAnswering.from_pretrained(model_path)

def ask(question: str, context: str) -> str:
    inputs = tokenizer_new(question, context, max_length=384,
        truncation="only_second", padding="max_length", return_tensors="pt")
    with torch.no_grad():
        outputs = model_new(**inputs)
    
    answer_start_index = outputs.start_logits.argmax()
    answer_end_index = outputs.end_logits.argmax()

    predict_answer_tokens = inputs.input_ids[0, answer_start_index : answer_end_index + 1]
    answer = tokenizer_new.decode(predict_answer_tokens)
    return answer
    return f"Question: '{question}'\nAnswer: {answer}"

# print(ask('What God created at first', 'Genesis 1:1 In the beginning God created the heaven and the earth.'))
# Streamlit App
st.set_page_config(page_title="Bible Q&A Bot", page_icon="📖", layout="centered")

st.title("📖 Bible Q&A Bot")
st.markdown("## Ask any question about the Bible and get scripturally grounded answers.")
st.write("‼️Only english🇺🇸 language provided‼️")


# User input
query = st.text_input("Enter your question:")

clear_button = st.button("Clear")
if clear_button:
    for key in st.session_state.keys():
        del st.session_state[key]

st.markdown('### Choose option to provide context')
option = st.radio("Choose how to provide context:", ("Manually", "Select Bible Verse"), label_visibility="collapsed")

def print_answer(question, context, answer):
    if context.isascii() and question.isascii():
        st.markdown("### ❓Question❓")
        st.write(question)
        st.markdown("### 📖Context📖")
        st.write(context)
        st.markdown("### ✅Answer✅")
        st.write(answer)
    else:
        st.error("Please ensure both the question and context are in English.")


import pandas as pd

# bible = pd.read_json("bible-dpo.json")

bible = pd.read_json("hf://datasets/nbeerbower/bible-dpo/bible-dpo.json")

books = list(bible['book'].unique())
v_by_b_c = bible.groupby(by=['book', 'chapter']).size().to_dict()

if option == "Manually":
    context = st.text_area("Enter the context (optional):")
    submit_button = st.button("Get Answer")

    if submit_button and query:
        with st.spinner("Searching Scripture..."):
            answer = ask(query, context)

        print_answer(query, context, answer)
    

elif option == "Select Bible Verse":


    book_name = st.selectbox("Select the book name:", [""] + books)

    if book_name:
        max_chapter = len(bible[bible["book"] == book_name].groupby('chapter'))
        chapter = st.selectbox("Select the chapter:", [""] + list(range(1, max_chapter + 1)))
    else:
        chapter = st.selectbox("Select the chapter:", [""])

    if chapter:
        max_verse = v_by_b_c.get((book_name, int(chapter)), 0)
        verse = st.selectbox("Enter the verse:", [""] + list(range(1, max_verse + 1)))
    else:
        verse = st.selectbox("Enter the verse:", [""])

    fetch_context_button = st.button("Fetch Verse")

    if query and fetch_context_button and book_name and chapter and verse:
        context = bible[
            (bible["book"] == book_name) &
            (bible['chapter'] == chapter) &
            (bible['verse'] == verse)
        ]['text'].values[0]
        with st.spinner("Searching Scripture..."):
            answer = ask(query, context)

        print_answer(query, context, answer)