app.py

import gradio as gr
from transformers import T5ForConditionalGeneration, T5Tokenizer

# Load the T5 model and tokenizer for question generation
model_name = "valhalla/t5-small-qg-prepend"
tokenizer = T5Tokenizer.from_pretrained(model_name)
model = T5ForConditionalGeneration.from_pretrained(model_name)

def generate_questions(email_text):
    # Prepend "generate questions: " to the input text
    input_text = "generate questions: " + email_text
    input_ids = tokenizer.encode(input_text, return_tensors="pt")

    # Generate questions
    outputs = model.generate(
        input_ids=input_ids,
        max_length=512,
        num_beams=4,
        early_stopping=True
    )

    # Decode the generated text
    questions = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return questions

# Create a Gradio interface
iface = gr.Interface(
    fn=generate_questions,
    inputs="textbox",
    outputs="textbox",
    title="Email Question Generator",
    description="Input an email, and the AI will generate the biggest questions that probably need to be answered.",
)

# Launch the interface
iface.launch()