app.py

import streamlit as st
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import torch
from examples import dialogue_examples

def generate_summary(model, tokenizer, dialogue):
    # Tokenize input dialogue
    inputs = tokenizer(dialogue, return_tensors="pt", max_length=1024, truncation=True)

    # Generate summary
    with torch.no_grad():
        summary_ids = model.generate(inputs["input_ids"], max_length=150, length_penalty=0.8, num_beams=4)

    # Decode and return the summary
    summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True, clean_up_tokenization_spaces=True)
    return summary

st.set_page_config(
    page_title="Dialogue Summarizer App",
    page_icon="ale.png",  # You can set your own emoji or use an image URL
)

#logo_path = "ale.png"
#logo_html = f'<div style="text-align:center;"><img src="{logo_path}" width="200"></div>'
#st.markdown(logo_html, unsafe_allow_html=True)

# Display the app name below the logo
st.title("Dialogue Summarizer App")

st.info("\n🖥️ Note: This application is running on CPU. Please be patient ⏳.")
st.markdown("This app summarizes dialogues. Enter a short dialogue in the text area. For best results, keep the dialogues at least a few sentences. You can also use the examples provided at the bottom of the page.")

# Create two columns layout using st.columns
col1, col2 = st.columns(2)

# User input on the left side with increased height
user_input = col1.text_area("Enter a Dialogue:", height=300)

# Summary textbox on the right side with initial value (read-only)
initial_summary = "Generated Summary will appear here."
generated_summary = col2.text_area("Summary:", value=initial_summary, height=300, key="summary")

# Add "Summarize" and "Clear" buttons
summarize_button = col1.button("Summarize")

# If "Summarize" button is clicked and there is user input, generate and display summary in the summary textbox
if summarize_button and user_input:
    # Load pre-trained Pegasus model and tokenizer
    model_name = "ale-dp/pegasus-finetuned-dialog-summarizer"
    tokenizer = AutoTokenizer.from_pretrained("google/pegasus-cnn_dailymail")
    model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

    # Generate summary
    summary = generate_summary(model, tokenizer, user_input)

    # Update the summary textbox with the generated summary
    generated_summary.text(summary)

st.markdown("**Dialogue examples:**")
for idx, example in enumerate(dialogue_examples, 1):
    st.write(f"Example {idx}:\n{example}")