app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
import torch

# Load grammar correction model
grammar_model_name = "prithivida/grammar_error_correcter_v1"
grammar_tokenizer = AutoTokenizer.from_pretrained(grammar_model_name)
grammar_model = AutoModelForSeq2SeqLM.from_pretrained(grammar_model_name)

# Load summarization pipeline
summarizer = pipeline("summarization", model="facebook/bart-large-cnn")

# Load title generation model (T5)
title_model_name = "t5-base"
title_tokenizer = AutoTokenizer.from_pretrained(title_model_name)
title_model = AutoModelForSeq2SeqLM.from_pretrained(title_model_name)

def polish_abstract(text):
    # --- 1. Grammar Correction ---
    grammar_input = grammar_tokenizer("gec: " + text, return_tensors="pt", max_length=512, truncation=True)
    grammar_output = grammar_model.generate(**grammar_input, max_length=512)
    corrected_text = grammar_tokenizer.decode(grammar_output[0], skip_special_tokens=True)

    # --- 2. Summary Generation ---
    summary_text = summarizer(corrected_text, max_length=120, min_length=40, do_sample=False)[0]['summary_text']

    # --- 3. Title Generation ---
    title_prompt = "generate title: " + corrected_text
    title_inputs = title_tokenizer(title_prompt, return_tensors="pt", max_length=512, truncation=True)
    title_output = title_model.generate(**title_inputs, max_length=20, num_beams=5)
    generated_title = title_tokenizer.decode(title_output[0], skip_special_tokens=True)

    return corrected_text, summary_text, generated_title

# Gradio interface
iface = gr.Interface(
    fn=polish_abstract,
    inputs=gr.Textbox(lines=12, label="Paste Your Abstract"),
    outputs=[
        gr.Textbox(label="✅ Grammar-Corrected Abstract"),
        gr.Textbox(label="🪄 Concise Summary"),
        gr.Textbox(label="📘 AI-Suggested Title"),
    ],
    title="🧠 AI Abstract & Title Polisher",
    description="This tool corrects grammar, summarizes, and suggests a title for biomedical abstracts."
)

iface.launch()