app.py

import os
import nltk 
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import torch
import streamlit as st

from src.doc2vec import inference
from src.abstractive_sum import summarize_text_with_model
from src.textrank import custom_textrank_summarizer
from src.clean import clean_license_text

CUSTOM_MODEL_NAME = "utkarshsaboo45/ClearlyDefinedLicenseSummarizer"

nltk.download('punkt')
os.environ["TOKENIZERS_PARALLELISM"] = "false"
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")

with st.spinner('Loading...'):
    model = AutoModelForSeq2SeqLM.from_pretrained(CUSTOM_MODEL_NAME).to(device)
    tokenizer = AutoTokenizer.from_pretrained(CUSTOM_MODEL_NAME)

summarization_type = st.sidebar.selectbox(
    "Select summarization type.",
    ("Abstractive", "Extractive", "Both")
)
if summarization_type == 'Abstractive':
    st.sidebar.caption('Summary will be generated by the T5 Transformer Model')
elif summarization_type == 'Extractive':
    st.sidebar.caption('Summary will be generated by a custom TextRank Algorithm')
    summary_len = st.sidebar.slider('Summary length percentage', 1, 10, 3)
elif summarization_type == 'Both':
    st.sidebar.caption('The License text will be first passed through the custom TextRank algorithm and then passed on to the T5 Transformer Model to generate a summary.')

clean_text = st.sidebar.checkbox('Show cleaned license text')

st.title('Clearly Defined: License Summarizer')
input = st.text_area('Enter contents of the license')

if len(input) > 0:
    with st.spinner('Loading...'):
        if summarization_type == 'Abstractive':
            summary, definitions = summarize_text_with_model(input, model, tokenizer)
        if summarization_type == 'Extractive':
                summary, definitions = custom_textrank_summarizer(input, summary_len = summary_len/10)
        if summarization_type == 'Both':
            summary, definitions = summarize_text_with_model(input, model, tokenizer)
            summary, _ = custom_textrank_summarizer(summary, summary_len = 1)
        
        if clean_text:
            st.header('Cleaned License Text')
            st.write(clean_license_text(input)[0])

        st.header('Summary')
        st.write(summary)
        
        prediction_scores = inference(input)
        st.header('Similarity Index')
        st.dataframe(prediction_scores)

        if definitions:
            st.header('Definitions')
            st.write(definitions)