import streamlit as st
import pandas as pd
from plms.language_model import TransformersQG
import time
import os
import numpy as np
st.set_page_config(page_icon='🧪', page_title='ViQAG for Vietnamese Education', layout='centered', initial_sidebar_state="collapsed")
with open(r"./static/styles.css") as f:
st.markdown(f"", unsafe_allow_html=True)
st.markdown(f"""
""", unsafe_allow_html=True)
st.markdown("AlphaEdu
", unsafe_allow_html=True)
# =====================================================================================================
if 'output' not in st.session_state:
st.session_state.output = ''
def file_selector(folder_path=r'./Resources/'):
filenames = os.listdir(folder_path)
return filenames
filenames = file_selector()
def load_grades(file_name, folder_path=r'./Resources/'):
file_path = f"{folder_path}{file_name}"
df = pd.read_csv(file_path)
list_grades = df['grade'].drop_duplicates().values
return list_grades, df
def load_chapters(df, grade_name):
df_raw = df[df['grade'] == grade_name]
list_chapters = df_raw['chapter'].drop_duplicates().values
return list_chapters, df
def load_lessons(df, grade_name, chapter_name):
df_raw = df[(df['grade'] == grade_name) & (df['chapter'] == chapter_name)]
return df_raw['lesson'].drop_duplicates().values
def load_context(df, grade_name, chapter_name, lesson_name):
context = df[(df['grade'] == grade_name) & (df['chapter'] == chapter_name) & (df['lesson'] == lesson_name)]['context'].values
return len(context), context
def generateQA(context, model_path = 'shnl/vit5-vinewsqa-qg-ae'):
unique_qa_pairs = set()
model = TransformersQG(model=model_path, max_length=512)
output = model.generate_qa(context)
qa_pairs = ''
for item in output:
question, answer = item
if (question, answer) not in unique_qa_pairs:
qa_pairs += f'question: {question} \nanswer: {answer} [SEP] '
unique_qa_pairs.add((question, answer))
qa = '\n\n'.join(qa_pairs.split(' [SEP] '))
return qa
# =====================================================================================================
col_1, col_2 = st.sidebar.columns(spec=[1, 1])
subject = col_1.selectbox(label='Select your subject:', options=filenames, label_visibility='visible')
list_grades, df = load_grades(file_name=subject)
grade = col_2.selectbox(label='Select your grade:', options=list_grades, label_visibility='visible')
list_chapters, df = load_chapters(df=df, grade_name=grade)
chapter = st.sidebar.selectbox(label='Select your chapter:', options=list_chapters, label_visibility='visible')
lesson_names = load_lessons(df=df, grade_name=grade, chapter_name=chapter)
lesson = st.sidebar.selectbox(label='Lesson:', options=lesson_names, label_visibility='visible')
total_paragraph, context_values = load_context(df=df, grade_name=grade, chapter_name=chapter, lesson_name=lesson)
col_12, col_22 = st.sidebar.columns(spec=[4, 6])
paragraph_idx = col_12.selectbox(label='Paragraph:', options=list(np.arange(1, total_paragraph + 1)), label_visibility='visible')
paragraph = st.text_area(label='Paragraph content', label_visibility='visible', height=200, value=context_values[paragraph_idx - 1])
col_22.selectbox(label='QAG model:', options=['ViT5-ViNewsQA'], label_visibility='visible')
btn_show_answer = st.sidebar.toggle(label='Show answers', disabled=False)
col_14, col_24, col_34, col_44, col_54 = st.columns(spec=[1, 1, 1, 1, 1])
btn_generate = col_34.button(label='Generate', use_container_width=True)
if btn_generate == True:
with st.spinner(text='Generating QA pairs from the selected paragraph. Please wait ...'):
st.session_state.output = generateQA(context=paragraph)
if btn_show_answer:
if st.session_state.output != '':
st.markdown("Generated QA pairs:", unsafe_allow_html=True)
st.code(body=st.session_state.output, language='latex')
else:
pass
else:
if st.session_state.output != '':
st.markdown("Generated QA pairs:", unsafe_allow_html=True)
output_no_answer = st.session_state.output.split(' [SEP] ')[0].split(', answer: ')[0].replace('question: ', '')
st.code(body=output_no_answer, language='latex')
else:
pass